D 2 : Decentralized Training over Decentralized Data

Size: px
Start display at page:

Download "D 2 : Decentralized Training over Decentralized Data"

Transcription

1 D : Deceralzed rag over Deceralzed Daa Hal ag Xagru La Mg Ya 3 Ce Zhag 4 J Lu 5 Absrac Whle rag a mache learg model usg mulple workers, each of whch collecs daa from s ow daa source, would be useful whe he daa colleced from dffere workers are uque ad dffere. Irocally, rece aalyss of deceralzed parallel sochasc grade desce D-PSGD reles o he assumpo ha he daa hosed o dffere workers are o oo dffere. I hs paper, we ask he queso: Ca we desg a deceralzed parallel sochasc grade desce algorhm ha s less sesve o he daa varace across workers? I hs paper, we prese D, a ovel deceralzed parallel sochasc grade desce algorhm desged for large daa varace amog workers mprecsely, deceralzed daa. he core of D s a varace reduco exeso of D-PSGD. I mproves he σ covergece rae from O + ζ 3 o /3 σ O where ζ deoes he varace amog daa o dffere workers. As a resul, D s robus o daa varace amog workers. We emprcally evaluaed D o mage classfcao asks, where each worker has access o oly he daa of a lmed se of labels, ad fd ha D sgfcaly ouperforms D-PSGD.. Iroduco rag mache learg models a deceralzed way has araced esve eress recely La e al., 07a; * Equal corbuo Deparme of Compuao Scece, Uversy of Rocheser Deparme of Compuaoal Mahemacs, Scece ad Egeerg, Mchga Sae Uversy 3 Deparme of Mahemacs, Mchga Sae Uversy 4 Deparme of Compuer Scece, EH Zurch 5 ece AI Lab. Correspodece o: Hal ag <hag4@ur.rocheser.edu>, Xagru La <xagru@yadex.com>, Mg Ya <yam@mah.msu.edu>, Ce Zhag <ce.zhag@f.ehz.ch>, J Lu <j.lu.uwsc@gmal.com>. Proceedgs of he 35 h Ieraoal Coferece o Mache Learg, Sockholm, Swede, PMLR 80, 08. Copyrgh 08 by he auhors. Yua e al., 06; Col e al., 06. I he deceralzed seg, here s a se of workers, each of whch collecs daa from dffere daa sources. Isead of sedg all daa o a ceralzed place, hese workers oly commucae wh her eghbors. he goal s o ge a model ha s he same as f all daa are colleced a ceralzed place. Deceralzed learg algorhms are mpora scearos where he ceralzed commucao s expesve or mpossble, or he uderlyg commucao ework has hgh laecy. For deceralzed learg o provde beefs, each user should provde daa ha s somehow uque,.e., he varace of daa colleced from dffere workers are large. However, may rece heorecal resuls La e al., 07a;b; Nedc & Ozdaglar, 009; Yua e al., 06 assume a bouded daa varace across workers whe daa hosed o dffere workers are very dffere, hese approaches coverge slowly, boh emprcally ad heorecally. I hs paper, we am a brgg hs dscrepacy bewee he curre heorecal udersadg ad he requremes from some praccal scearos. I hs paper, we prese D, a ovel deceralzed learg algorhm desged o be robus uder hgh daa varace. D s bul upo deceralzed parallel sochasc grade desce D-PSGD, bu beefs from a addoal varace reduco compoe. I D, each worker sores he sochasc grade ad s local model he prevous erae ad learly combes hem wh he curre sochasc grade ad local model. I resuls a mproved covergece rae over D-PSGD by elmag he daa varao amog workers. I parcular, he covergece rae s mproved from σ O + ζ 3 o O σ /3 where ζ s he daa varao amog all workers, σ s he daa varace wh each worker, s he umber of workers, ad s he umber of eraos. We emprcally show D ca sgfcaly ouperform D-PSGD by rag a mage classfcao model where each worker has access o oly he daa of a lmed se of labels. hroughou hs paper, we cosder he followg deceralzed opmzao: m x R N fx:= {}}{ E ξ D F x; ξ, =:f x

2 Deceralzed rag over Deceralzed Daa where s he umber of workers ad D s he local daa dsrbuo for worker. All workers are coeced hrough a coeced graph. Each worker ca oly exchage formao wh s eghbors. Defos ad oao hroughou hs paper, we use followg oao ad defos: F deoes he Frobeus orm of marces. deoes he l orm for vecors ad he specral orm for marces. f deoes he grade of a fuco f. f deoes he opmal soluo of. λ deoes he h larges egevalue of a marx. x deoes he local model of worker. F x ; ξ deoes a local sochasc grade of worker. = [,,, ] R deoes he all-oe vecor. I order o orgaze he algorhm more clearly, here we defe he cocaeao of all local varables, sochasc grades, ad her averages respecvely: X :=[x,..., x ] R N, X :=X = x, GX; ξ :=[ F x ; ξ,..., F x ; ξ ] R N, GX, ξ :=GX, ξ = fx := f X, fx := f x, F x ; ξ, where ξ s he colleco of radomly sampled daa from all workers. Orgazao hs paper s orgazed as follows: Seco revews relaed work abou he proposed approach; Seco 3 roduces he sae-of-he-ar deceralzed sochasc grade desce mehod ad s covergece rae; Seco 4 roduces he proposed algorhm ad s uo why mproves he sae-of-he-ar approach; Seco 5 provdes he heorecal guaraee; ad Seco 6 valdaes he proposed approaches va emprcal sudy; ad Seco 7 cocludes hs paper.. Relaed work I hs seco, we revew he sochasc grade desce algorhm ad s deceralzed varas, deceralzed algorhms, ad prevous varace reduco echologes. Sochasc grade desce SGD he SGD approahces Ghadm & La, 03; Moules & Bach, 0; Nemrovsk e al., 009 s que powerful for solvg largescale mache learg problems. I acheves a covergece rae of O /. As a mplemeao of SGD, he Ceralzed Parallel Sochasc Grade Desce C-PSGD, has bee wdely used parallel compuao. I C-PSGD, a ceral worker, whose job s o perform he varable updaes, s coeced o may leaf workers ha are used o compue sochasc grades parallel. C-PSGD has bee appled o may deep learg frameworks, such as CNK Sede & Agarwal, 06, MXNe Che e al., 05, ad esorflow Abad e al., 06. he covergece rae of C- PSGD s O, whch shows ha ca acheve lear speedup wh regards o he umber of leaf workers. Deceralzed algorhms Ceralzed algorhms requre a ceral server o commucae wh all oher workers Suresh e al., 07. I coras, deceralzed algorhms work o ay coeced ework ad oly rely o he formao exchage bewee eghbor workers Kashyap e al., 007; Lavae & Murray, 0; Nedc e al., 009. Deceralzed algorhms are especally useful uder a ework wh lmed badwdh or hgh laecy. I s more favorable whe daa prvacy s sesve. hese advaages have led o successful applcaos. he deceralzed approach for mul-ask reforceme learg was suded Omdshafe e al. 07; Mhamd e al. 07. I Col e al. 06, a dual based deceralzed algorhm was proposed o solve he parwse fuco opmzao. Sh e al. 04 ad Mokhar & Rbero 05 aalyzed he deceralzed verso of he ADMM opmzao algorhm. A formao heorec approach was used o aalyze deceralzao Dobbe e al. 07. he deceralzed verso of sub-grade desce was suded Nedc & Ozdaglar 009; Yua e al. 06. Is O/ covergece requres a dmshg sepsze or a cosa sepsze ha depeds o he oal umber of eraos. hs pheomeo happes because of he varace bewee he daa dffere workers, whch we call ouer varace o dffereae from he varace SGD. Recely, here are several deermsc deceralzed opmzao algorhms ha allows a cosa sepsze. For example, EXRA Sh e al., 05a s he frs modfcao of deceralzed grade desce ha coverges uder a cosa sepsze. Laer hs algorhm s exeded for problems wh he sum of smooh ad osmooh fucos a each ode Sh e al., 05b.

3 Deceralzed rag over Deceralzed Daa he algorhm DIGg s proposed Nedć e al. 07, where wo exchages are eeded each erao. However, her sepszes deped o boh he Lpschz cosa of he dffereable fuco ad he ework srucure. NIDS s he frs algorhm ha has a cosa ework depede sepsze L e al., 07. hs algorhm was smulaeously proposed by Yua e al. 07 for he smooh case oly usg a dffere approach. Deceralzed parallel sochasc grade desce D- PSGD he D-PSGD algorhm Nedc & Ozdaglar, 009; Ram e al., 00a;b requres each worker o compue a sochasc grade ad exchage s local model wh eghbors. I Duch e al. 0, a dual averagg based mehod s proposed for solvg he cosraed deceralzed SGD opmzao. I Yua e al. 06, he covergece rae for D-PSGD was aalyzed whe he grade s assumed o be bouded. I La e al. 07, a deceralzed prmal-dual ype mehod was proposed wh a compuaoal complexy of O /ɛ for geeral covex objecves. La e al. 07a proved ha D-PSGD ca adms lear speedup wh respec o he umber of workers wh a smlar covergece rae as C-PSGD. Varace reduco echology here have bee may mehods developed for reducg he varace SGD, cludg SVRG Johso & Zhag, 03, SAGA Defazo e al., 04, SAG Schmd e al., 07, MISO Maral, 05, ad msgd Koečỳ e al., 06. However, mos of hese echologes are desged for ceralzed approaches. he DSA algorhm Mokhar & Rbero, 06 appled he varace reduco smlar o SAGA o srogly covex deceralzed opmzao problems ad proved a lear covergece rae. However, he speedup propery s uclear ad a able of all sochasc grades eed o be sored. 3. Prelmary: deceralzed sochasc grade desce he deceralzed sochasc grade desce La e al., 07a; Zhag e al., 07; Shahrampour & Jadbabae, 07 allows each worker say worker maag s ow local varable x. Durg each erao say, erao, each worker performs he followg seps:. Query s eghbors local varables.. ake weghed average wh s local varable ad s eghbors local varables: x = W + j x j, j= where W j s he, j eleme of he marx W. W j = 0 meas worker ad worker j are o coeced. 3. Perform oe sochasc grade desce sep x + = x + γ F x ; ξ, where ξ represes he daa sampled worker a he erao followg he dsrbuo D. From a global po of vew, he updae rule of D-PSGD ca be vewed as X + = X W γgx ; ξ. I adms he followg rae show heorem. heorem Covergece rae of D-PSGD La e al., 07a. Uder cera assumpos, he oupu of D-PSGD adms he followg equaly γl =0 f0 f γ E fx + D E f X =0 + γl σ + γ L σ λd + 9γ L ς λ D, where ρ reflecs he propery of he ework, D ad D are defed o be D := D := 9γ L ρ, D, 8γ ρ L ad σ ad ς measure he varao wh each worker ad amog all workers respecvely E ξ D F x; ξ f x σ,, x, f x fx ζ,, x. 3 Choosg he opmal seplegh γ = L+σ K + 3 ζ 3 3 we have he followg covergece rae: E fx σ O + 3 ζ = he proposed D algorhm ca mprove he covergece rae by removg he depedece o he global boud of ouer varace ζ.

4 Deceralzed rag over Deceralzed Daa Algorhm he D algorhm : Ipu: Ial po x 0 = 0, sep legh γ > 0, cofuso marx W, ad he oal umber of eraos. : for = 0,,,..., do from he local daa of he h worker. 4: Compue a local sochasc grade based o ξ ad 3: Radomly sample ξ curre varable x 5: f =0 he 6: x + 7: else 8: x + = x = x : F x ; ξ. γ F x ; ξ, x γ F x ; ξ + γ F x ; ξ. 9: ed f 0: Each worker seds x o s eghbors ad akes + he weghed average x + = j= W j x j, + where x j s from he worker j. + : ed for : Oupu: x 4. he D algorhm I D algorhm, each worker say, worker repeas he followg updag rule say, a erao :. Compue a local sochasc grade F x samplg ξ from dsrbuo D ;. Updae he local model x γ F x ; ξ + + γ F x ; ξ ; ξ by x x usg he local models ad sochasc grades boh he h erao ad he h erao. 3. Whe he sychrozao barrer s me, exchage wh eghbors: x + x + = j= W j x j. + From a global po of vew, he updae rule of D ca be vewed as: X + = X X γgx ; ξ + γgx ; ξ W. he complee algorhm s summarzed Algorhm. D esseally rus he sochasc grade desce sep. o udersad he uo of D, le us cosder he mea value X, whch s updaed jus lke he sadard sochasc grade desce: X + = X X γgx ; ξ + γgx ; ξ W, X + =X X γgx ; ξ + γgx ; ξ, or equvalely X + X =X X γgx ; ξ + γgx ; ξ, =X X 0 γ GX ; ξ GX ; ξ k= = γgx ; ξ. X = X 0 γgx 0 ; ξ 0. 4 Why D mproves he D-PSGD? We may oce ha D- PSGD also esseally updaes he form of sochasc grade desce 4. he why D ca mprove D-PSGD? Assume ha X has acheved he opmum X := x wh all local models equal o he opmum x o. he for D-PSGD, he ex updae wll be X + = X γgx ; ξ. I shows ha he covergece whe we approach a soluo s affeced by E[ GX ; ξ F ], whch s bouded by Oσ + ζ, as we ca see from he followg: E[ GX ; ξ F ] =E F x ; ξ + f x + f x E F x ; ξ + f x + f x fx σ + ζ. Nex we apply a smlar aalyss for D by assumg ha boh X ad X have reached he opmal soluo X. he ex updae for D wll be: X + = X γgx ; ξ γgx ; ξ W. I shows ha for D, he covergece whe we approach a soluo reles o he magude of E[ GX ; ξ GX ; ξ F ], whch s bouded by: Oσ,

5 Deceralzed rag over Deceralzed Daa whch ca be see from: E[ GX ; ξ GX ; ξ F =E F x ; ξ f x E σ. F x ; ξ f x 5. heorecal guaraee hs seco provdes he heorecal guaraee for he proposed D algorhm. We frs gve he assumpos requred below. Assumpo. hroughou hs paper, we make he followg commoly used assumpos:. Lpschza grade: All fuco f s are wh L-Lpschza grades.. Bouded varace: Assume bouded varace of sochasc grade wh each worker E ξ D F x; ξ f x σ,, x. 3. Symmerc cofuso marx: he cofuso marx W s symmerc ad sasfes W =. 4. Specral gap: Le he egevalues of W R be λ λ λ. Deoe by for shor λ := max λ = λ. {,,} We assume λ < ad λ > Ialzao: W.l.o.g., assume all local varables are alzed by zero, ha s, X 0 = 0. Exsg deceralzed cosesus algorhms Sh e al., 05b; L e al., 07 use a modfcao of he doubly sochasc marx such ha λ > 0,.e., choose W = W + I/ where W s a doubly sochasc marx. Recely, L & Ya 07 show ha λ > /3 s opmal he covergece of EXRA. However, he opmal λ for NIDS L e al., 07 s ukow. I hs paper, we proved ha 3 s he fmum of λ, ad whe reduces o deermsc case, hs codo s weaker ha ha L e al., 07. hs s mpora, because we acually ca use a W ha performs beer. Gve Assumpo, we have followg covergece guaraee for D : heorem Covergece of Algorhm. Choose he seplegh γ Algorhm o be a cosa sasfyg 4C γ L > 0. Uder Assumpo, we have he followg covergece rae for Algorhm : A f0 + E fx + A E fx f0 f γ + L C γ σ where ζ 0 := = + L γ σ + 6L C γ ζ 0 + 6L C γ 4 L σ + 6L C γ σ, f 0 f0, v :=λ λ λ, { } C := max v, λ, { λ C := max v, λ }, λ λ := 4C γ L, A := 6L C γ, A := Lγ 6L C γ 4 L. By appropraely specfyg he sep legh γ, we reach he followg corollary: Corollary 3. Choose he sep legh γ Algorhm o be γ = 8 C, where C ad C are defed L+6 C L+σ heorem. Uder Assumpo, he followg covergece rae holds E fx σ + + ζ0 + σ =0 + σ + σ, where ζ 0 s defed heorem ad we rea f0 f, L, λ, ad λ as cosas. Noe ha we ca oba eve beer cosas by choosg dffere parameers ad applyg gher equales, however, he ma resul of hs corollary s o show he order of he covergece. We hghlgh a few key observaos from our heorecal resuls he followg. ghess of he covergece rae Seg σ = 0 ad ζ 0 = 0, whch reduces he VR-SGD o a ormal GD 5

6 Deceralzed rag over Deceralzed Daa algorhm, we shall see ha he covergece rae becomes O, whch s exacly he rae of GD. Lear speedup Sce he leadg erm of he covergece rae s O, whch s cosse wh he covergece rae of C-PSGD, hs dcaes ha we would acheve a lear speed up wh respec o he umber of odes. Cosse wh NIDS I NIDS L & Ya, 07, he ζ 0 erm depeds o ζ 0 he covergece rae s O. Whle he correspodg erm D ζ s O 0 +σ, whch dcaes whe our algorhm s cosse wh NIDS because NIDS σ s cosdered o be 0. Superory over D-PSGD Whe compared o D-PSGD, he covergece rae of D oly depeds o ζ 0, ad he correspodg decayg rae s ζ0. Whereas D-PSGD La e al., 07a, we eed o assume a upper boud for he global varace bewee dffere odes daase, ad s fluece ca be compared o σ, he er varace of each ode self. hs meas we ca always acheve a much beer covergece rae ha D-PSGD. 6. Expermes We evaluae he effecveess of D by comparg wh boh ceralzed ad deceralzed SGD algorhms. 6.. Experme Segs We coduc expermes wo segs.. RANSFERLEARNING: We es he case ha each worker has access o a local pre-raed eural ework as feaure exracor, ad we wa o ra a logsc regresso model amog all hese workers. I our experme, we selec he frs 6 classes of ImageNe ad use IcepoV4 as he feaure exracor o exrac 048 feaures for each mage. We coduc daa augmeao ad geerae a blurblack verso for each mage. I oal hs daase coas mages.. LENE: We es he case ha all workers collaboravely ra a eural ework model. We ra a LeNe o he CIFAR0 daase. I oal hs daase coas 50,000 mages of sze 3 3. Oe cavea of rag more rece eural eworks s ha moder archecures ofe have a bach ormalzao layer, whch herely assumes ha he daa dsrbuo s uform across dffere baches, whch s o he case ha we are eresed. I prcple, we could also flow he bach formao hrough he ework a deceralzed way; however, we leave hs as fuure work. By defaul, each worker oly has exclusve access o a subse of classes. For RANSFERLEARNING, we use 6 workers ad each worker has access o oe class; for LENE, we use 5 workers ad each worker has access o wo classes. For comparso, we also cosder a case whe he daases s frs shuffled ad he uformly paroed amog all he workers, we call hs he shuffled case, ad he defaul oe he ushuffled case. We use a rg opology for boh expermes. Parameer ug. For RANSFERLEARNING, we use cosa learg raes ad ue from {0.0, 0.05, 0.05, 0.075, 0.}. For LENE, we use cosa learg rae 0.05 whch s ued from {0.5, 0., 0.05, 0.0} for ceralzed algorhms ad bach sze 8 o each worker. Mercs. I hs paper, we maly focus o he covergece rae of dffere algorhms sead of he wall clock speed. hs s because he mplemeao of D s a mor chage over he sadard D-PSGD algorhm, ad hus hey has almos he same speed o fsh oe epoch of rag, ad boh are o slower ha he ceralzed algorhm. Whe he ework has hgh laecy, f a deceralzed algorhm D or D-PSGD coverges wh a smlar speed as he ceralzed algorhm, ca be up o oe order of magude faser La e al., 07a. However, he covergece rae depedg o he ouer varace s dffere for boh algorhms. 6.. Ushuffled Case varao across workers s maxmzed. Fgure shows he resul. I he ushuffled case, we see ha D-PSGD coverges slower ha he ceralzed case. hs s cosse wh he orgal D-PSGD paper La e al., 07a. O he oher had, D coverges much faser ha D-PSGD, ad acheves almos he same loss as he ceralzed algorhm. For he LeNe case, each worker oly has access o daa of assged wo labels, whch meas he daa varao s very large. he D-PSGD does o coverge wh he gve learg rae Shuffled Case As a say check, Fgure shows he resul of hree dffere algorhms o he shuffled daa. I hs case, he daa varao amog workers s small expecao, hey are draw from he same dsrbuo. We see ha, all sraeges have smlar covergece rae. hs valdae ha D s more effecve for larger daa varao bewee dffere workers. We ca ue he learg rae 50x smaller for D-PSGD o coverge hs case, bu dog so wll make D-PSGD suck a he sarg po for que a log me.

7 Deceralzed rag over Deceralzed Daa Loss Deceralzed D Ceralzed # Epochs a RANSFERLEARNING Loss D Deceralzed Ceralzed # Epochs b LENE Fgure. Covergece of Dffere Dsrbued rag Algorhms Ushuffled Case..5.5 Deceralzed Loss 0.5 Loss D Ceralzed 0 Deceralzed D Ceralzed # Epochs # Epochs a RANSFERLEARNING b LENE Fgure. Covergece of Dffere Dsrbued rag Algorhms Shuffled Case. 7. Cocluso I hs paper, we propose a deceralzed algorhm, amely, D algorhm. D algorhm egraes he D-PSGD algorhm wh he varace reduco echology, by whch we mproves he covergece rae of D-PSGD. he varace reduco echology used hs paper s dffere from he commoly used oes such as SVRG ad SAGA, ha are desged for ceralzed approaches. Expermes valdae he advaage of D over D-PSGD D coverges wh a rae ha s smlar o ceralzed SGD whle D-PSGD does o coverge o a soluo wh a smlar qualy whe he daa varace s large. Whle beg robus o large daa varace amog workers, he same performace beef of D-PSGD over he ceralzed sraegy sll holds for D. Ackowledgemes hs projec s suppored par by NSF CCF7853, NEC fellowshp, IBM faculy award, NSF DMS-6798, Swss NSF NRP , IBM Zurch, Mercedes- Bez Research & Developme Norh Amerca, Oracle

8 Deceralzed rag over Deceralzed Daa Labs, Swsscom, Zurch Isurace, ad Chese Scholarshp Coucl. Refereces Abad, M., Barham, P., Che, J., Che, Z., Davs, A., Dea, J., Dev, M., Ghemawa, S., Irvg, G., Isard, M., e al. esorflow: A sysem for large-scale mache learg. I OSDI, volume 6, pp , 06. Che,., L, M., L, Y., L, M., Wag, N., Wag, M., Xao,., Xu, B., Zhag, C., ad Zhag, Z. Mxe: A flexble ad effce mache learg lbrary for heerogeeous dsrbued sysems. arxv prepr arxv:5.074, 05. Col, I., Belle, A., Salmo, J., ad Clémeço, S. Gossp dual averagg for deceralzed opmzao of parwse fucos. I Ieraoal Coferece o Mache Learg, pp , 06. Defazo, A., Bach, F., ad Lacose-Jule, S. Saga: A fas cremeal grade mehod wh suppor for osrogly covex compose objecves. I Advaces eural formao processg sysems, pp , 04. Dobbe, R., Frdovch-Kel, D., ad oml, C. Fully deceralzed polces for mul-age sysems: A formao heorec approach. I Advaces Neural Iformao Processg Sysems, pp , 07. Duch, J. C., Agarwal, A., ad Wawrgh, M. J. Dual averagg for dsrbued opmzao: Covergece aalyss ad ework scalg. IEEE rasacos o Auomac corol, 573:59 606, 0. Ghadm, S. ad La, G. Sochasc frs- ad zeroh-order mehods for ocovex sochasc programmg. SIAM Joural o Opmzao, 34:34 368, 03. do: 0.37/ Johso, R. ad Zhag,. Accelerag sochasc grade desce usg predcve varace reduco. I Advaces eural formao processg sysems, pp , 03. Kashyap, A., Başar,., ad Srka, R. Quazed cosesus. Auomaca, 437:9 03, 007. Koečỳ, J., Lu, J., Rchárk, P., ad akáč, M. M-bach sem-sochasc grade desce he proxmal seg. IEEE Joural of Seleced opcs Sgal Processg, 0 :4 55, 06. La, G., Lee, S., ad Zhou, Y. Commucao-effce algorhms for deceralzed ad sochasc opmzao Lavae, J. ad Murray, R. M. Quazed cosesus by meas of gossp algorhm. IEEE rasacos o Auomac Corol, 57:9 3, 0. L, Z. ad Ya, M. A prmal-dual algorhm wh opmal sepszes ad s applcao deceralzed cosesus opmzao. arxv prepr arxv: , 07. L, Z., Sh, W., ad Ya, M. A deceralzed proxmalgrade mehod wh ework depede sep-szes ad separaed covergece raes. arxv prepr arxv: , 07. La, X., Zhag, C., Zhag, H., Hseh, C.-J., Zhag, W., ad Lu, J. Ca deceralzed algorhms ouperform ceralzed algorhms? a case sudy for deceralzed parallel sochasc grade desce a. La, X., Zhag, W., Zhag, C., ad Lu, J. Asychroous deceralzed parallel sochasc grade desce. arxv prepr arxv: , 07b. Maral, J. Icremeal majorzao-mmzao opmzao wh applcao o large-scale mache learg. SIAM Joural o Opmzao, 5:89 855, 05. Mhamd, E., Mahd, E., Hedrkx, H., Guerraou, R., ad Maurer, A. D. O. Dyamc safe errupbly for deceralzed mul-age reforceme learg. echcal repor, EPFL, 07. Mokhar, A. ad Rbero, A. Deceralzed double sochasc averagg grade. I Sgals, Sysems ad Compuers, 05 49h Aslomar Coferece o, pp IEEE, 05. Mokhar, A. ad Rbero, A. Dsa: Deceralzed double sochasc averagg grade algorhm. Joural of Mache Learg Research, 76: 35, 06. Moules, E. ad Bach, F. R. No-asympoc aalyss of sochasc approxmao algorhms for mache learg. I Shawe-aylor, J., Zemel, R. S., Barle, P. L., Perera, F., ad Weberger, K. Q. eds., Advaces Neural Iformao Processg Sysems 4, pp Curra Assocaes, Ic., 0. Nedc, A. ad Ozdaglar, A. Dsrbued subgrade mehods for mul-age opmzao. IEEE rasacos o Auomac Corol, 54:48 6, 009. Nedc, A., Olshevsky, A., Ozdaglar, A., ad sskls, J. N. O dsrbued averagg algorhms ad quazao effecs. IEEE rasacos o Auomac Corol, 54: , 009. Nedć, A., Olshevsky, A., ad Rabba, M. G. Nework opology ad commucao-compuao radeoffs deceralzed opmzao. arxv prepr arxv: , 07.

9 Deceralzed rag over Deceralzed Daa Nemrovsk, A., Judsky, A., La, G., ad Shapro, A. Robus sochasc approxmao approach o sochasc programmg. SIAM Joural o Opmzao, 94: , 009. do: 0.37/ Omdshafe, S., Pazs, J., Amao, C., How, J. P., ad Va, J. Deep deceralzed mul-ask mul-age rl uder paral observably. arxv prepr arxv: , 07. Ram, S. S., Nedć, A., ad Veeravall, V. V. Asychroous gossp algorhm for sochasc opmzao: Cosa sepsze aalyss. I Rece Advaces Opmzao ad s Applcaos Egeerg, pp Sprger, 00a. Yua, K., Lg, Q., ad Y, W. O he covergece of deceralzed grade desce. SIAM Joural o Opmzao, 63: , 06. do: 0.37/ Yua, K., Yg, B., Zhao, X., ad Sayed, A. H. Exac dffuso for dsrbued opmzao ad learg par : Algorhm developme. arxv prepr arxv:70.05, 07. Zhag, W., Zhao, P., Zhu, W., Ho, S. C., ad Zhag,. Projeco-free dsrbued ole learg eworks. I Ieraoal Coferece o Mache Learg, pp , 07. Ram, S. S., Nedć, A., ad Veeravall, V. V. Dsrbued sochasc subgrade projeco algorhms for covex opmzao. Joural of opmzao heory ad applcaos, 473:56 545, 00b. Schmd, M., Le Roux, N., ad Bach, F. Mmzg fe sums wh he sochasc average grade. Mahemacal Programmg, 6-:83, 07. Sede, F. ad Agarwal, A. Ck: Mcrosof s ope-source deep-learg oolk. I Proceedgs of he Nd ACM SIGKDD Ieraoal Coferece o Kowledge Dscovery ad Daa Mg, KDD 6, pp , New York, NY, USA, 06. ACM. ISBN do: 0.45/ Shahrampour, S. ad Jadbabae, A. Dsrbued ole opmzao dyamc evromes usg mrror desce. IEEE rasacos o Auomac Corol, 07. Sh, W., Lg, Q., Yua, K., Wu, G., ad Y, W. O he lear covergece of he admm deceralzed cosesus opmzao. IEEE ras. Sgal Processg, 67: , 04. Sh, W., Lg, Q., Wu, G., ad Y, W. Exra: A exac frsorder algorhm for deceralzed cosesus opmzao. SIAM Joural o Opmzao, 5: , 05a. Sh, W., Lg, Q., Wu, G., ad Y, W. A proxmal grade algorhm for deceralzed compose opmzao. IEEE rasacos o Sgal Processg, 63: , 05b. Suresh, A.., Yu, F. X., Kumar, S., ad McMaha, H. B. Dsrbued mea esmao wh lmed commucao. I Precup, D. ad eh, Y. W. eds., Proceedgs of he 34h Ieraoal Coferece o Mache Learg, volume 70 of Proceedgs of Mache Learg Research, pp , Ieraoal Coveo Cere, Sydey, Ausrala, 06 Aug 07. PMLR.

10 Deceralzed rag over Deceralzed Daa Supplemeal Maerals hs suppleme maeral cludes he proofs for heorem. Because he cofuso marx W s symmerc, ca be decomposed as W = P ΛP, where P = v, v,, v s a orhogoal marx,.e., P P = P P = I, ad Λ = dag{λ,..., λ } s a dagoal marx wh dagoal eres beg he egevalues of W he ocreasg order. he applyg he decomposo o he erao from W ad W o W + gves X + = X W X W γgx ; ξ W + γgx ; ξ W X + =X P ΛP X P ΛP γgx ; ξ P ΛP + γgx ; ξ P ΛP. Deoe Y = X P, HX ; ξ = GX ; ξ P, ad use y respecvely. he or he colums of Y ad HX ; ξ, ad h o dcae he -h colum of Y ad HX ; ξ, Y + =Y Λ Y Λ γhx ; ξ Λ + γhx ; ξ Λ, 6 y + =λ y y γh + γh. 7 From he properes of W Assumpo ad he decomposo, we have λ = ad v =,,,. herefore y = X. For all oher egevalues 3 < λ <, he equao 7 shows ha all y explas how he cofuso marx works. Lemma 4. Gve wo o-egave sequeces {a } = ad {b } = ha sasfyg wh ρ [0,, we have Proof. S k := D k := a = = a = = = = S k := D k := a = would decay o zero, whch ρ s b s, 8 a = b s ρ, a ρ = ρ s b s = ρ s b s = r= ρ where he las equaly holds because of 9. = =s ρ r b r = r= ρ s r b s + b r ρ s b s b s. ρ s b s = k s ρ b s =0 = r= = ρ = r= ρ s r b s b r ρ s r b s b s ρ. 9 b s 0

11 Deceralzed rag over Deceralzed Daa Lemma 5. For ay marx X R N, we have X v X v = X F = X P e = X P F = X F where e R wh he -h compoe beg ad all ohers beg 0. Proof. From he defo of he Frobeus orm for a marx, we have X v = X P F = r X P P X = r X X = X F. Sce X v 0, so X v X v = = X F. I he same way, we have X P e = X P = X F F. he resul s proved. Lemma 6. Gve ρ 3, 0 0,, for ay wo sequece {a } =0 ad {b } =0 ha sasfy a 0 =b 0 = 0, a =b, a + =ρa a + b b,, we have u + v + a + = a u v + u s+ v s+ β s, 0, u v where β s = b s b s, u = ρ + ρ ρ, v = ρ ρ ρ. More specfcally, f 0 < ρ <, we have a + s θ = a ρ / s [ + θ] + β s ρ s/ s [ + sθ], 0 where β s =b s b s, θ = arccos ρ. Proof. Whe = 0, he resuls s easy o verfy. Nex we cosder he case. Sce a + = ρa ρa + β,

12 Deceralzed rag over Deceralzed Daa We ca fd u = ρ + ρ ρ, v = ρ ρ ρ, such ha a + ua =a ua v + β. Noe ha u ad v are complex umbers whe 0 < ρ <. ha s wh θ = arccos ρ. Recursvely applyg gves Dvg boh sdes by u +, we oba u = ρe θ, v = ρe θ, a + ua =a ua v + β = a ua v + β v + β =a ua 0 v + β s v s a + u + = a u + u + = a + u u = a u + =a v + β s v s. due o a 0 = 0 a v + β s v s a v + β s v s + u a + v + u a k v k + k= he we mulply boh sdes by u + ad have a + =a u + u a k v k + k= =a u + β s v k s β s v k s v k + u u k= v k =a u + u u k=0 k=s v + =a u u v + u u u + v + =a + u v k= β s v s v u β s v s v u β s v s v k u k due o s v u u s+ v s+ β s. u v Whe ρ 0,, sce u = ρe θ ad v = ρe θ, we have k= s+ v u β s v s a s b k = k=s a s b k he resul s proved. / s [ + θ] a + = a ρ s θ + s/ s [ s + θ] β s ρ. s θ

13 Lemma 7. Uder Assumpo, we have 4C γ L Deceralzed rag over Deceralzed Daa =0 X x C X F + C γ σ + 6C γ 4 L σ + 6C γ 4 L fx, where γ, L, σ, θ, C ad C are defed heorem. Proof. o esmae he dfferece of he local models ad he global mea model, we have X x = X e X = X X F = X P P X v v F 0, 0, 0,, 0 0,, 0,, 0 = X P 0, 0,,, , 0, 0,, F = y, where y s he -h colum of X P. Noe ha we have, from 7, where β = γh = y + = λ y y γh + γh = λ y y + λ β, + γh. For all y ha correspodg o 3 < λ < 0, Lemma 6 shows y u + + =y v + + λ u v β s where u = λ + λ λ ad v = λ λ λ. herefore, we have For u+ v + u v y +, we have Usg 3, we oba u + y u + v + u v + λ u s+ β s = v s+, u v u s+ v s+ u v v + u u v v u v v u v v due o u < v. y + Summg from = 0 o = gves y + =0 = y = y v + λ y β s v s. v + λ =0 = β s. 3 v s.

14 Deoe a = have β s where v = λ λ λ. Deceralzed rag over Deceralzed Daa v s, whch has he same srucure as he sequece Lemma 4. herefore, whe λ < 0, we y = y v + λ v = y v + λ v For all y ha sasfes 0 λ <, from 7 ad Lemma 6, we have where β s he = γh s y + y + s θ = y λ/ s [ + θ ] + λ + γh s ad θ = arccos λ. s θ y y Summg from = 0 o gves y + =0 s θ = = β β, 4 β s λ s/ s [ + sθ ], λ s [ + θ ] + λ β s s [ + sθ ] λ s/ λ + λ y = s θ β s λ s/, y =0 λ + λ = β s λ s/ From Lemma 4, he s θ = λ gves { Deoe C = max β s λ s/ v, y = y = λ } has he same srucure as he sequece Lemma 4, so we have y s θ + λ λ λ β = y λ + λ λ λ = y λ + λ λ λ ad C = max y = =. β β. 5 { } λ v, λ. From 4 ad 5, we have λ λ y C + C β =. 6

15 Deceralzed rag over Deceralzed Daa We ex boud β = E = = β γ E h h =γ E G X ; ξ P e G X ; ξ P e γ = E G X ; ξ P e G X ; ξ P e =γ E G X ; ξ P G X ; ξ P F =γ E G X ; ξ G X ; ξ F due o Lemma 5 =γ E F x ; ξ F x ; ξ =γ F E =3γ E F + 3γ f 6γ σ + 3γ 6γ σ + 3γ x x x ; ξ f x F x ; ξ f x + f ; ξ f x f x E f L E x x + 3γ f x x =6γ σ + 3γ L E Y P e Y P e F x ; ξ f x x f x =6γ σ + 3γ L E Y P Y P F =6γ σ + 3γ L E Y Y F due o Lemma 5 =6γ σ + 3γ L E y y. 7 Combg 6 ad 7, we have = = y C Y F + C β = = C Y F + C 6γ σ + 3γ L = C Y F + C γ σ + 6C γ L E y E y = y y. 8

16 he ex sep s o boud E y y. Because Deceralzed rag over Deceralzed Daa y = X P e = X v = X = X, wha we eed o boud s E X + X. From 4, we have X + = X γg. herefore E X+ X = γ E G = γ E G fx + γ fx γ σ ad we have he follow boud for E y y Combg 8 ad 9 we ge 4C γ L = = = = y y E ogeher wh ad X 0 = 0, we have 4C γ L = : y + y + γ fx, γ σ + γ fx. 9 C Y F + C γ σ + 6C γ 4 L σ + 6C γ 4 L fx + 6C γ L E = = y y C Y F + C γ σ + 6C γ 4 L σ + 6C γ 4 L fx y + 6C γ L E + = = y C Y F + C γ σ + 6C γ 4 L σ + 6C γ 4 L fx y + 6C γ L E + = = y = = = due o y 0 = 0 C Y F + C γ σ + 6C γ 4 L σ + 6C γ 4 L fx + 4C γ L E = = y, C Y F + C γ σ + 6C γ 4 L σ + 6C γ 4 L fx. X x C Y F + C γ σ + 6C γ 4 L σ + 6C γ 4 L fx due o X F = Y F Acually, whe λ 3, we have v, he y ad C X F + C γ σ + 6C γ 4 L σ + 6C γ 4 L fx. = X x. he algorhm would fal o coverge hs suao, ad hs s why /3 s he fmum of λ. = = = =

17 Lemma 8. Followg he Assumpo, we have EfX + EfX γ E fx Proof. From 4, we have Deceralzed rag over Deceralzed Daa γ Lγ E fx + γ E fx fx + Lγ σ. X + = X γ GX ; ξ. From em of Assumpo, we kow ha f has a L-Lpschz couous grade. So, we have EfX + EfX + E fx, γ GX ; ξ + L E γ GX ; ξ =EfX + E fx, γ E ξ GX ; ξ + Lγ E GX ; ξ =EfX γ E fx, fx + Lγ E GX ; ξ fx + fx =EfX γ E fx, fx + Lγ E GX ; ξ fx + Lγ E fx + Lγ E E ξ GX ; ξ fx, fx =EfX γ E fx, fx + Lγ E GX ; ξ fx + Lγ E fx =EfX γ E fx, fx + Lγ E F x ; ξ f x + Lγ E fx =EfX γ E fx, fx + Lγ + E ξ F x E E F x ; ξ f x ; ξ f x, E ξ F x ; ξ f x EfX γ E fx, fx + Lγ σ + Lγ E fx + Lγ E fx =EfX γ E fx γ E fx + γ E fx fx + Lγ E fx + Lγ σ due o a, b = a + b a b =EfX γ E fx γ Lγ E fx + γ E fx fx + Lγ σ, 0 whch complees he proof. Proof o heorem Proof. We frs esmae he upper boud for E fx fx : E fx fx = E L f X f x E f X f x E X x.

18 Combg 0 Lemma 8 ad yelds Seg γ = γ, we oba Deceralzed rag over Deceralzed Daa γ E fx + E fx + Lγ E fx γ From Lemma 7, we have γ Lγ E fx EfX EfX + + γ E fx fx + Lγ EfX EfX + + L γ X x + Lγ σ. 4C γ L =0 σ EfX f EfX + f + L X x C X F + C γ σ + 6C γ 4 L σ + 6C γ 4 L fx, If γ s o oo large ha sasfes 4C γ L > 0, he deoe = 4C γ L, we would have =0 X x C X F + C γ σ Summarzg boh sdes of ad applyg 3 yelds I mples E fx + Lγ E fx =0 EfX 0 f γ f0 f γ =0 + 6L C γ 4 L + L =0 + L γ σ + L C fx. = = + 6C γ 4 L σ E X x + L γ σ X F + L C γ σ E fx + Lγ 6L C γ 4 L E fx f0 f γ = f0 f γ + L γ σ + L C + L γ σ + L C γ X F + L C γ σ G0; ξ 0 F + L C γ σ + 6C γ 4 L X x + Lγ σ. fx. 3 = + 6L C γ 4 L σ + 6L C γ 4 L σ However, G0; ξ 0 F ca be expaded as: G0, ξ 0 F = F 0, ξ f 0 + f 0 f0 + f0 + 6L C γ 4 L σ. 4 3σ + 3ζ f0, 5

19 Deceralzed rag over Deceralzed Daa where ζ 0 = f 0 f0 dcaes he dfferece bewee dffere workers daase a he sar po. Combg 4 ad 5, he we have he we have Deoe becomes 6L C γ f0 f γ E fx + Lγ 6L C γ 4 L E fx =0 f0 f γ + L γ σ + L C γ σ + 6L C γ 4 L σ + 6L C γ σ + 6L C γ ζ 0 + 6L C γ f0. f0 + = + L γ σ + L C γ σ E fx + A = 6L C γ Lγ 6L C γ 4 L E fx + 6L C γ 4 L σ + 6L C γ σ + 6L C γ ζ 0. A = Lγ 6L C γ 4 L, A f0 + E fx + A E fx f0 f γ I complees he proof. = + L γ σ + L C γ σ + 6L C γ 4 L σ + 6L C γ σ + 6L C γ ζ 0. Proof o Corollary 3 Proof. From he value of γ, we oba herefore C γ L 64, C γ L 36. = 4C γ L, A = 6L C γ, A = Lγ 6L C γ 4 L > 0, γ L + σ, γ 4 L 4 + σ 4.

20 Deceralzed rag over Deceralzed Daa he we ca remove he fx ad f0 o he lef had sde of 5 heorem, ad 5 becomes whch complees he proof. E fx 4f0 f L8 C + 6 C =0 + 4f0 f σ + Lσ + 48L C σ L + σ + 4L4 σ C L 4 + σ 4 + 4L C σ L + σ + 4L C ζ 0 L + σ,

AN INCREMENTAL QUASI-NEWTON METHOD WITH A LOCAL SUPERLINEAR CONVERGENCE RATE. Aryan Mokhtari Mark Eisen Alejandro Ribeiro

AN INCREMENTAL QUASI-NEWTON METHOD WITH A LOCAL SUPERLINEAR CONVERGENCE RATE. Aryan Mokhtari Mark Eisen Alejandro Ribeiro AN INCREMENTAL QUASI-NEWTON METHOD WITH A LOCAL SUPERLINEAR CONVERGENCE RATE Arya Mokhar Mark Ese Alejadro Rbero Deparme of Elecrcal ad Sysems Egeerg, Uversy of Pesylvaa ABSTRACT We prese a cremeal Broyde-Flecher-Goldfarb-Shao

More information

VARIATIONAL ITERATION METHOD FOR DELAY DIFFERENTIAL-ALGEBRAIC EQUATIONS. Hunan , China,

VARIATIONAL ITERATION METHOD FOR DELAY DIFFERENTIAL-ALGEBRAIC EQUATIONS. Hunan , China, Mahemacal ad Compuaoal Applcaos Vol. 5 No. 5 pp. 834-839. Assocao for Scefc Research VARIATIONAL ITERATION METHOD FOR DELAY DIFFERENTIAL-ALGEBRAIC EQUATIONS Hoglag Lu Aguo Xao Yogxag Zhao School of Mahemacs

More information

Key words: Fractional difference equation, oscillatory solutions,

Key words: Fractional difference equation, oscillatory solutions, OSCILLATION PROPERTIES OF SOLUTIONS OF FRACTIONAL DIFFERENCE EQUATIONS Musafa BAYRAM * ad Ayd SECER * Deparme of Compuer Egeerg, Isabul Gelsm Uversy Deparme of Mahemacal Egeerg, Yldz Techcal Uversy * Correspodg

More information

The Poisson Process Properties of the Poisson Process

The Poisson Process Properties of the Poisson Process Posso Processes Summary The Posso Process Properes of he Posso Process Ierarrval mes Memoryless propery ad he resdual lfeme paradox Superposo of Posso processes Radom seleco of Posso Pos Bulk Arrvals ad

More information

The Linear Regression Of Weighted Segments

The Linear Regression Of Weighted Segments The Lear Regresso Of Weghed Segmes George Dael Maeescu Absrac. We proposed a regresso model where he depede varable s made o up of pos bu segmes. Ths suao correspods o he markes hroughou he da are observed

More information

Solution set Stat 471/Spring 06. Homework 2

Solution set Stat 471/Spring 06. Homework 2 oluo se a 47/prg 06 Homework a Whe he upper ragular elemes are suppressed due o smmer b Le Y Y Y Y A weep o he frs colum o oba: A ˆ b chagg he oao eg ad ec YY weep o he secod colum o oba: Aˆ YY weep o

More information

QR factorization. Let P 1, P 2, P n-1, be matrices such that Pn 1Pn 2... PPA

QR factorization. Let P 1, P 2, P n-1, be matrices such that Pn 1Pn 2... PPA QR facorzao Ay x real marx ca be wre as AQR, where Q s orhogoal ad R s upper ragular. To oba Q ad R, we use he Householder rasformao as follows: Le P, P, P -, be marces such ha P P... PPA ( R s upper ragular.

More information

-distributed random variables consisting of n samples each. Determine the asymptotic confidence intervals for

-distributed random variables consisting of n samples each. Determine the asymptotic confidence intervals for Assgme Sepha Brumme Ocober 8h, 003 9 h semeser, 70544 PREFACE I 004, I ed o sped wo semesers o a sudy abroad as a posgraduae exchage sude a he Uversy of Techology Sydey, Ausrala. Each opporuy o ehace my

More information

Fundamentals of Speech Recognition Suggested Project The Hidden Markov Model

Fundamentals of Speech Recognition Suggested Project The Hidden Markov Model . Projec Iroduco Fudameals of Speech Recogo Suggesed Projec The Hdde Markov Model For hs projec, s proposed ha you desg ad mpleme a hdde Markov model (HMM) ha opmally maches he behavor of a se of rag sequeces

More information

Determination of Antoine Equation Parameters. December 4, 2012 PreFEED Corporation Yoshio Kumagae. Introduction

Determination of Antoine Equation Parameters. December 4, 2012 PreFEED Corporation Yoshio Kumagae. Introduction refeed Soluos for R&D o Desg Deermao of oe Equao arameers Soluos for R&D o Desg December 4, 0 refeed orporao Yosho Kumagae refeed Iroduco hyscal propery daa s exremely mpora for performg process desg ad

More information

The Mean Residual Lifetime of (n k + 1)-out-of-n Systems in Discrete Setting

The Mean Residual Lifetime of (n k + 1)-out-of-n Systems in Discrete Setting Appled Mahemacs 4 5 466-477 Publshed Ole February 4 (hp//wwwscrporg/oural/am hp//dxdoorg/436/am45346 The Mea Resdual Lfeme of ( + -ou-of- Sysems Dscree Seg Maryam Torab Sahboom Deparme of Sascs Scece ad

More information

8. Queueing systems lect08.ppt S Introduction to Teletraffic Theory - Fall

8. Queueing systems lect08.ppt S Introduction to Teletraffic Theory - Fall 8. Queueg sysems lec8. S-38.45 - Iroduco o Teleraffc Theory - Fall 8. Queueg sysems Coes Refresher: Smle eleraffc model M/M/ server wag laces M/M/ servers wag laces 8. Queueg sysems Smle eleraffc model

More information

Least squares and motion. Nuno Vasconcelos ECE Department, UCSD

Least squares and motion. Nuno Vasconcelos ECE Department, UCSD Leas squares ad moo uo Vascocelos ECE Deparme UCSD Pla for oda oda we wll dscuss moo esmao hs s eresg wo was moo s ver useful as a cue for recogo segmeao compresso ec. s a grea eample of leas squares problem

More information

The algebraic immunity of a class of correlation immune H Boolean functions

The algebraic immunity of a class of correlation immune H Boolean functions Ieraoal Coferece o Advaced Elecroc Scece ad Techology (AEST 06) The algebrac mmuy of a class of correlao mmue H Boolea fucos a Jgla Huag ad Zhuo Wag School of Elecrcal Egeerg Norhwes Uversy for Naoales

More information

arxiv: v1 [stat.ml] 21 Mar 2017

arxiv: v1 [stat.ml] 21 Mar 2017 Sochasc Prmal Dual Coordae Mehod wh No-Uform Samplg Based o Opmaly Volaos Asush Shbagak shbagak.a.mllab.@gmal.com Deparme of Scefc ad Egeerg Smulao, Nagoya Isue of Techology arxv:703.0706v [sa.ml] Mar

More information

(1) Cov(, ) E[( E( ))( E( ))]

(1) Cov(, ) E[( E( ))( E( ))] Impac of Auocorrelao o OLS Esmaes ECON 3033/Evas Cosder a smple bvarae me-seres model of he form: y 0 x The four key assumpos abou ε hs model are ) E(ε ) = E[ε x ]=0 ) Var(ε ) =Var(ε x ) = ) Cov(ε, ε )

More information

Real-Time Systems. Example: scheduling using EDF. Feasibility analysis for EDF. Example: scheduling using EDF

Real-Time Systems. Example: scheduling using EDF. Feasibility analysis for EDF. Example: scheduling using EDF EDA/DIT6 Real-Tme Sysems, Chalmers/GU, 0/0 ecure # Updaed February, 0 Real-Tme Sysems Specfcao Problem: Assume a sysem wh asks accordg o he fgure below The mg properes of he asks are gve he able Ivesgae

More information

14. Poisson Processes

14. Poisson Processes 4. Posso Processes I Lecure 4 we roduced Posso arrvals as he lmg behavor of Bomal radom varables. Refer o Posso approxmao of Bomal radom varables. From he dscusso here see 4-6-4-8 Lecure 4 " arrvals occur

More information

The Bernstein Operational Matrix of Integration

The Bernstein Operational Matrix of Integration Appled Mahemacal Sceces, Vol. 3, 29, o. 49, 2427-2436 he Berse Operaoal Marx of Iegrao Am K. Sgh, Vee K. Sgh, Om P. Sgh Deparme of Appled Mahemacs Isue of echology, Baaras Hdu Uversy Varaas -225, Ida Asrac

More information

Partial Molar Properties of solutions

Partial Molar Properties of solutions Paral Molar Properes of soluos A soluo s a homogeeous mxure; ha s, a soluo s a oephase sysem wh more ha oe compoe. A homogeeous mxures of wo or more compoes he gas, lqud or sold phase The properes of a

More information

FALL HOMEWORK NO. 6 - SOLUTION Problem 1.: Use the Storage-Indication Method to route the Input hydrograph tabulated below.

FALL HOMEWORK NO. 6 - SOLUTION Problem 1.: Use the Storage-Indication Method to route the Input hydrograph tabulated below. Jorge A. Ramírez HOMEWORK NO. 6 - SOLUTION Problem 1.: Use he Sorage-Idcao Mehod o roue he Ipu hydrograph abulaed below. Tme (h) Ipu Hydrograph (m 3 /s) Tme (h) Ipu Hydrograph (m 3 /s) 0 0 90 450 6 50

More information

Fault Tolerant Computing. Fault Tolerant Computing CS 530 Probabilistic methods: overview

Fault Tolerant Computing. Fault Tolerant Computing CS 530 Probabilistic methods: overview Probably 1/19/ CS 53 Probablsc mehods: overvew Yashwa K. Malaya Colorado Sae Uversy 1 Probablsc Mehods: Overvew Cocree umbers presece of uceray Probably Dsjo eves Sascal depedece Radom varables ad dsrbuos

More information

Quantum Mechanics II Lecture 11 Time-dependent perturbation theory. Time-dependent perturbation theory (degenerate or non-degenerate starting state)

Quantum Mechanics II Lecture 11 Time-dependent perturbation theory. Time-dependent perturbation theory (degenerate or non-degenerate starting state) Pro. O. B. Wrgh, Auum Quaum Mechacs II Lecure Tme-depede perurbao heory Tme-depede perurbao heory (degeerae or o-degeerae sarg sae) Cosder a sgle parcle whch, s uperurbed codo wh Hamloa H, ca exs a superposo

More information

Midterm Exam. Tuesday, September hour, 15 minutes

Midterm Exam. Tuesday, September hour, 15 minutes Ecoomcs of Growh, ECON560 Sa Fracsco Sae Uvers Mchael Bar Fall 203 Mderm Exam Tuesda, Sepember 24 hour, 5 mues Name: Isrucos. Ths s closed boo, closed oes exam. 2. No calculaors of a d are allowed. 3.

More information

New Guaranteed H Performance State Estimation for Delayed Neural Networks

New Guaranteed H Performance State Estimation for Delayed Neural Networks Ieraoal Joural of Iformao ad Elecrocs Egeerg Vol. o. 6 ovember ew Guaraeed H Performace ae Esmao for Delayed eural eworks Wo Il Lee ad PooGyeo Park Absrac I hs paper a ew guaraeed performace sae esmao

More information

As evident from the full-sample-model, we continue to assume that individual errors are identically and

As evident from the full-sample-model, we continue to assume that individual errors are identically and Maxmum Lkelhood smao Greee Ch.4; App. R scrp modsa, modsb If we feel safe makg assumpos o he sascal dsrbuo of he error erm, Maxmum Lkelhood smao (ML) s a aracve alerave o Leas Squares for lear regresso

More information

FORCED VIBRATION of MDOF SYSTEMS

FORCED VIBRATION of MDOF SYSTEMS FORCED VIBRAION of DOF SSES he respose of a N DOF sysem s govered by he marx equao of moo: ] u C] u K] u 1 h al codos u u0 ad u u 0. hs marx equao of moo represes a sysem of N smulaeous equaos u ad s me

More information

Continuous Time Markov Chains

Continuous Time Markov Chains Couous me Markov chas have seay sae probably soluos f a oly f hey are ergoc, us lke scree me Markov chas. Fg he seay sae probably vecor for a couous me Markov cha s o more ffcul ha s he scree me case,

More information

IMPROVED PORTFOLIO OPTIMIZATION MODEL WITH TRANSACTION COST AND MINIMAL TRANSACTION LOTS

IMPROVED PORTFOLIO OPTIMIZATION MODEL WITH TRANSACTION COST AND MINIMAL TRANSACTION LOTS Vol.7 No.4 (200) p73-78 Joural of Maageme Scece & Sascal Decso IMPROVED PORTFOLIO OPTIMIZATION MODEL WITH TRANSACTION COST AND MINIMAL TRANSACTION LOTS TIANXIANG YAO AND ZAIWU GONG College of Ecoomcs &

More information

Supplement Material for Inverse Probability Weighted Estimation of Local Average Treatment Effects: A Higher Order MSE Expansion

Supplement Material for Inverse Probability Weighted Estimation of Local Average Treatment Effects: A Higher Order MSE Expansion Suppleme Maeral for Iverse Probably Weged Esmao of Local Average Treame Effecs: A Hger Order MSE Expaso Sepe G. Doald Deparme of Ecoomcs Uversy of Texas a Aus Yu-C Hsu Isue of Ecoomcs Academa Sca Rober

More information

Some Probability Inequalities for Quadratic Forms of Negatively Dependent Subgaussian Random Variables

Some Probability Inequalities for Quadratic Forms of Negatively Dependent Subgaussian Random Variables Joural of Sceces Islamc epublc of Ira 6(: 63-67 (005 Uvers of ehra ISSN 06-04 hp://scecesuacr Some Probabl Iequales for Quadrac Forms of Negavel Depede Subgaussa adom Varables M Am A ozorga ad H Zare 3

More information

For the plane motion of a rigid body, an additional equation is needed to specify the state of rotation of the body.

For the plane motion of a rigid body, an additional equation is needed to specify the state of rotation of the body. The kecs of rgd bodes reas he relaoshps bewee he exeral forces acg o a body ad he correspodg raslaoal ad roaoal moos of he body. he kecs of he parcle, we foud ha wo force equaos of moo were requred o defe

More information

AML710 CAD LECTURE 12 CUBIC SPLINE CURVES. Cubic Splines Matrix formulation Normalised cubic splines Alternate end conditions Parabolic blending

AML710 CAD LECTURE 12 CUBIC SPLINE CURVES. Cubic Splines Matrix formulation Normalised cubic splines Alternate end conditions Parabolic blending CUIC SLINE CURVES Cubc Sples Marx formulao Normalsed cubc sples Alerae ed codos arabolc bledg AML7 CAD LECTURE CUIC SLINE The ame sple comes from he physcal srume sple drafsme use o produce curves A geeral

More information

International Journal Of Engineering And Computer Science ISSN: Volume 5 Issue 12 Dec. 2016, Page No.

International Journal Of Engineering And Computer Science ISSN: Volume 5 Issue 12 Dec. 2016, Page No. www.jecs. Ieraoal Joural Of Egeerg Ad Compuer Scece ISSN: 19-74 Volume 5 Issue 1 Dec. 16, Page No. 196-1974 Sofware Relably Model whe mulple errors occur a a me cludg a faul correco process K. Harshchadra

More information

Cyclone. Anti-cyclone

Cyclone. Anti-cyclone Adveco Cycloe A-cycloe Lorez (963) Low dmesoal aracors. Uclear f hey are a good aalogy o he rue clmae sysem, bu hey have some appealg characerscs. Dscusso Is he al codo balaced? Is here a al adjusme

More information

Average Consensus in Networks of Multi-Agent with Multiple Time-Varying Delays

Average Consensus in Networks of Multi-Agent with Multiple Time-Varying Delays I. J. Commucaos ewor ad Sysem Sceces 3 96-3 do:.436/jcs..38 Publshed Ole February (hp://www.scrp.org/joural/jcs/). Average Cosesus ewors of Mul-Age wh Mulple me-varyg Delays echeg ZHAG Hu YU Isue of olear

More information

Solving fuzzy linear programming problems with piecewise linear membership functions by the determination of a crisp maximizing decision

Solving fuzzy linear programming problems with piecewise linear membership functions by the determination of a crisp maximizing decision Frs Jo Cogress o Fuzzy ad Iellge Sysems Ferdows Uversy of Mashhad Ira 9-3 Aug 7 Iellge Sysems Scefc Socey of Ira Solvg fuzzy lear programmg problems wh pecewse lear membershp fucos by he deermao of a crsp

More information

Linear Regression Linear Regression with Shrinkage

Linear Regression Linear Regression with Shrinkage Lear Regresso Lear Regresso h Shrkage Iroduco Regresso meas predcg a couous (usuall scalar oupu from a vecor of couous pus (feaures x. Example: Predcg vehcle fuel effcec (mpg from 8 arbues: Lear Regresso

More information

Stability Criterion for BAM Neural Networks of Neutral- Type with Interval Time-Varying Delays

Stability Criterion for BAM Neural Networks of Neutral- Type with Interval Time-Varying Delays Avalable ole a www.scecedrec.com Proceda Egeerg 5 (0) 86 80 Advaced Corol Egeergad Iformao Scece Sably Crero for BAM Neural Neworks of Neural- ype wh Ierval me-varyg Delays Guoqua Lu a* Smo X. Yag ab a

More information

Solution. The straightforward approach is surprisingly difficult because one has to be careful about the limits.

Solution. The straightforward approach is surprisingly difficult because one has to be careful about the limits. ose ad Varably Homewor # (8), aswers Q: Power spera of some smple oses A Posso ose A Posso ose () s a sequee of dela-fuo pulses, eah ourrg depedely, a some rae r (More formally, s a sum of pulses of wdh

More information

Chapter 8. Simple Linear Regression

Chapter 8. Simple Linear Regression Chaper 8. Smple Lear Regresso Regresso aalyss: regresso aalyss s a sascal mehodology o esmae he relaoshp of a respose varable o a se of predcor varable. whe here s jus oe predcor varable, we wll use smple

More information

Available online Journal of Scientific and Engineering Research, 2014, 1(1): Research Article

Available online  Journal of Scientific and Engineering Research, 2014, 1(1): Research Article Avalable ole wwwjsaercom Joural o Scec ad Egeerg Research, 0, ():0-9 Research Arcle ISSN: 39-630 CODEN(USA): JSERBR NEW INFORMATION INEUALITIES ON DIFFERENCE OF GENERALIZED DIVERGENCES AND ITS APPLICATION

More information

EE 6885 Statistical Pattern Recognition

EE 6885 Statistical Pattern Recognition EE 6885 Sascal Paer Recogo Fall 005 Prof. Shh-Fu Chag hp://.ee.columba.edu/~sfchag Lecure 8 (/8/05 8- Readg Feaure Dmeso Reduco PCA, ICA, LDA, Chaper 3.8, 0.3 ICA Tuoral: Fal Exam Aapo Hyväre ad Erkk Oja,

More information

Fully Fuzzy Linear Systems Solving Using MOLP

Fully Fuzzy Linear Systems Solving Using MOLP World Appled Sceces Joural 12 (12): 2268-2273, 2011 ISSN 1818-4952 IDOSI Publcaos, 2011 Fully Fuzzy Lear Sysems Solvg Usg MOLP Tofgh Allahvraloo ad Nasser Mkaelvad Deparme of Mahemacs, Islamc Azad Uversy,

More information

The Optimal Combination Forecasting Based on ARIMA,VAR and SSM

The Optimal Combination Forecasting Based on ARIMA,VAR and SSM Advaces Compuer, Sgals ad Sysems (206) : 3-7 Clausus Scefc Press, Caada The Opmal Combao Forecasg Based o ARIMA,VAR ad SSM Bebe Che,a, Mgya Jag,b* School of Iformao Scece ad Egeerg, Shadog Uversy, Ja,

More information

Probability Bracket Notation and Probability Modeling. Xing M. Wang Sherman Visual Lab, Sunnyvale, CA 94087, USA. Abstract

Probability Bracket Notation and Probability Modeling. Xing M. Wang Sherman Visual Lab, Sunnyvale, CA 94087, USA. Abstract Probably Bracke Noao ad Probably Modelg Xg M. Wag Sherma Vsual Lab, Suyvale, CA 94087, USA Absrac Ispred by he Drac oao, a ew se of symbols, he Probably Bracke Noao (PBN) s proposed for probably modelg.

More information

Efficient Estimators for Population Variance using Auxiliary Information

Efficient Estimators for Population Variance using Auxiliary Information Global Joural of Mahemacal cece: Theor ad Praccal. IN 97-3 Volume 3, Number (), pp. 39-37 Ieraoal Reearch Publcao Houe hp://www.rphoue.com Effce Emaor for Populao Varace ug Aular Iformao ubhah Kumar Yadav

More information

The ray paths and travel times for multiple layers can be computed using ray-tracing, as demonstrated in Lab 3.

The ray paths and travel times for multiple layers can be computed using ray-tracing, as demonstrated in Lab 3. C. Trael me cures for mulple reflecors The ray pahs ad rael mes for mulple layers ca be compued usg ray-racg, as demosraed Lab. MATLAB scrp reflec_layers_.m performs smple ray racg. (m) ref(ms) ref(ms)

More information

Least Squares Fitting (LSQF) with a complicated function Theexampleswehavelookedatsofarhavebeenlinearintheparameters

Least Squares Fitting (LSQF) with a complicated function Theexampleswehavelookedatsofarhavebeenlinearintheparameters Leas Squares Fg LSQF wh a complcaed fuco Theeampleswehavelookedasofarhavebeelearheparameers ha we have bee rg o deerme e.g. slope, ercep. For he case where he fuco s lear he parameers we ca fd a aalc soluo

More information

Optimal Eye Movement Strategies in Visual Search (Supplement)

Optimal Eye Movement Strategies in Visual Search (Supplement) Opmal Eye Moveme Sraeges Vsual Search (Suppleme) Jr Naemk ad Wlso S. Gesler Ceer for Percepual Sysems ad Deparme of Psychology, Uversy of exas a Aus, Aus X 787 Here we derve he deal searcher for he case

More information

Redundancy System Fault Sampling Under Imperfect Maintenance

Redundancy System Fault Sampling Under Imperfect Maintenance A publcao of CHEMICAL EGIEERIG TRASACTIOS VOL. 33, 03 Gues Edors: Erco Zo, Pero Barald Copyrgh 03, AIDIC Servz S.r.l., ISB 978-88-95608-4-; ISS 974-979 The Iala Assocao of Chemcal Egeerg Ole a: www.adc./ce

More information

Asymptotic Regional Boundary Observer in Distributed Parameter Systems via Sensors Structures

Asymptotic Regional Boundary Observer in Distributed Parameter Systems via Sensors Structures Sesors,, 37-5 sesors ISSN 44-8 by MDPI hp://www.mdp.e/sesors Asympoc Regoal Boudary Observer Dsrbued Parameer Sysems va Sesors Srucures Raheam Al-Saphory Sysems Theory Laboraory, Uversy of Perpga, 5, aveue

More information

Real-time Classification of Large Data Sets using Binary Knapsack

Real-time Classification of Large Data Sets using Binary Knapsack Real-me Classfcao of Large Daa Ses usg Bary Kapsack Reao Bru bru@ds.uroma. Uversy of Roma La Sapeza AIRO 004-35h ANNUAL CONFERENCE OF THE ITALIAN OPERATIONS RESEARCH Sepember 7-0, 004, Lecce, Ialy Oule

More information

Continuous Indexed Variable Systems

Continuous Indexed Variable Systems Ieraoal Joural o Compuaoal cece ad Mahemacs. IN 0974-389 Volume 3, Number 4 (20), pp. 40-409 Ieraoal Research Publcao House hp://www.rphouse.com Couous Idexed Varable ysems. Pouhassa ad F. Mohammad ghjeh

More information

The textbook expresses the stock price as the present discounted value of the dividend paid and the price of the stock next period.

The textbook expresses the stock price as the present discounted value of the dividend paid and the price of the stock next period. ublc Affars 974 Meze D. Ch Fall Socal Sceces 748 Uversy of Wscos-Madso Sock rces, News ad he Effce Markes Hypohess (rev d //) The rese Value Model Approach o Asse rcg The exbook expresses he sock prce

More information

Mixed Integral Equation of Contact Problem in Position and Time

Mixed Integral Equation of Contact Problem in Position and Time Ieraoal Joural of Basc & Appled Sceces IJBAS-IJENS Vol: No: 3 ed Iegral Equao of Coac Problem Poso ad me. A. Abdou S. J. oaquel Deparme of ahemacs Faculy of Educao Aleadra Uversy Egyp Deparme of ahemacs

More information

Quantitative Portfolio Theory & Performance Analysis

Quantitative Portfolio Theory & Performance Analysis 550.447 Quaave Porfolo heory & Performace Aalyss Week February 4 203 Coceps. Assgme For February 4 (hs Week) ead: A&L Chaper Iroduco & Chaper (PF Maageme Evrome) Chaper 2 ( Coceps) Seco (Basc eur Calculaos)

More information

Research on portfolio model based on information entropy theory

Research on portfolio model based on information entropy theory Avalable ole www.jocpr.com Joural of Chemcal ad Pharmaceucal esearch, 204, 6(6):286-290 esearch Arcle ISSN : 0975-7384 CODEN(USA) : JCPC5 esearch o porfolo model based o formao eropy heory Zhag Jusha,

More information

arxiv: v2 [cs.lg] 19 Dec 2016

arxiv: v2 [cs.lg] 19 Dec 2016 1 Sasfcg mul-armed bad problems Paul Reverdy, Vabhav Srvasava, ad Naom Ehrch Leoard arxv:1512.07638v2 [cs.lg] 19 Dec 2016 Absrac Sasfcg s a relaxao of maxmzg ad allows for less rsky decso makg he face

More information

Moments of Order Statistics from Nonidentically Distributed Three Parameters Beta typei and Erlang Truncated Exponential Variables

Moments of Order Statistics from Nonidentically Distributed Three Parameters Beta typei and Erlang Truncated Exponential Variables Joural of Mahemacs ad Sascs 6 (4): 442-448, 200 SSN 549-3644 200 Scece Publcaos Momes of Order Sascs from Nodecally Dsrbued Three Parameers Bea ype ad Erlag Trucaed Expoeal Varables A.A. Jamoom ad Z.A.

More information

θ = θ Π Π Parametric counting process models θ θ θ Log-likelihood: Consider counting processes: Score functions:

θ = θ Π Π Parametric counting process models θ θ θ Log-likelihood: Consider counting processes: Score functions: Paramerc coug process models Cosder coug processes: N,,..., ha cou he occurreces of a eve of eres for dvduals Iesy processes: Lelhood λ ( ;,,..., N { } λ < Log-lelhood: l( log L( Score fucos: U ( l( log

More information

Asymptotic Behavior of Solutions of Nonlinear Delay Differential Equations With Impulse

Asymptotic Behavior of Solutions of Nonlinear Delay Differential Equations With Impulse P a g e Vol Issue7Ver,oveber Global Joural of Scece Froer Research Asypoc Behavor of Soluos of olear Delay Dffereal Equaos Wh Ipulse Zhag xog GJSFR Classfcao - F FOR 3 Absrac Ths paper sudes he asypoc

More information

4. THE DENSITY MATRIX

4. THE DENSITY MATRIX 4. THE DENSTY MATRX The desy marx or desy operaor s a alerae represeao of he sae of a quaum sysem for whch we have prevously used he wavefuco. Alhough descrbg a quaum sysem wh he desy marx s equvale o

More information

Fourth Order Runge-Kutta Method Based On Geometric Mean for Hybrid Fuzzy Initial Value Problems

Fourth Order Runge-Kutta Method Based On Geometric Mean for Hybrid Fuzzy Initial Value Problems IOSR Joural of Mahemacs (IOSR-JM) e-issn: 2278-5728, p-issn: 29-765X. Volume, Issue 2 Ver. II (Mar. - Apr. 27), PP 4-5 www.osrjourals.org Fourh Order Ruge-Kua Mehod Based O Geomerc Mea for Hybrd Fuzzy

More information

Lecture 3 Topic 2: Distributions, hypothesis testing, and sample size determination

Lecture 3 Topic 2: Distributions, hypothesis testing, and sample size determination Lecure 3 Topc : Drbuo, hypohe eg, ad ample ze deermao The Sude - drbuo Coder a repeaed drawg of ample of ze from a ormal drbuo of mea. For each ample, compue,,, ad aoher ac,, where: The ac he devao of

More information

COMPARISON OF ESTIMATORS OF PARAMETERS FOR THE RAYLEIGH DISTRIBUTION

COMPARISON OF ESTIMATORS OF PARAMETERS FOR THE RAYLEIGH DISTRIBUTION COMPARISON OF ESTIMATORS OF PARAMETERS FOR THE RAYLEIGH DISTRIBUTION Eldesoky E. Affy. Faculy of Eg. Shbee El kom Meoufa Uv. Key word : Raylegh dsrbuo, leas squares mehod, relave leas squares, leas absolue

More information

Bianchi Type II Stiff Fluid Tilted Cosmological Model in General Relativity

Bianchi Type II Stiff Fluid Tilted Cosmological Model in General Relativity Ieraoal Joural of Mahemacs esearch. IN 0976-50 Volume 6, Number (0), pp. 6-7 Ieraoal esearch Publcao House hp://www.rphouse.com Bach ype II ff Flud led Cosmologcal Model Geeral elay B. L. Meea Deparme

More information

The textbook expresses the stock price as the present discounted value of the dividend paid and the price of the stock next period.

The textbook expresses the stock price as the present discounted value of the dividend paid and the price of the stock next period. coomcs 435 Meze. Ch Fall 07 Socal Sceces 748 Uversy of Wscos-Madso Sock rces, News ad he ffce Markes Hypohess The rese Value Model Approach o Asse rcg The exbook expresses he sock prce as he prese dscoued

More information

Orbital Euclidean stability of the solutions of impulsive equations on the impulsive moments

Orbital Euclidean stability of the solutions of impulsive equations on the impulsive moments Pure ad Appled Mahemacs Joural 25 4(: -8 Publshed ole Jauary 23 25 (hp://wwwscecepublshggroupcom/j/pamj do: 648/jpamj254 ISSN: 2326-979 (Pr ISSN: 2326-982 (Ole Orbal ucldea sably of he soluos of mpulsve

More information

ENGINEERING solutions to decision-making problems are

ENGINEERING solutions to decision-making problems are 3788 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 62, NO. 8, AUGUST 2017 Sasfcg Mul-Armed Bad Problems Paul Reverdy, Member, IEEE, Vabhav Srvasava, ad Naom Ehrch Leoard, Fellow, IEEE Absrac Sasfcg s a

More information

Solution of Impulsive Differential Equations with Boundary Conditions in Terms of Integral Equations

Solution of Impulsive Differential Equations with Boundary Conditions in Terms of Integral Equations Joural of aheacs ad copuer Scece (4 39-38 Soluo of Ipulsve Dffereal Equaos wh Boudary Codos Ters of Iegral Equaos Arcle hsory: Receved Ocober 3 Acceped February 4 Avalable ole July 4 ohse Rabba Depare

More information

Density estimation III. Linear regression.

Density estimation III. Linear regression. Lecure 6 Mlos Hauskrec mlos@cs.p.eu 539 Seo Square Des esmao III. Lear regresso. Daa: Des esmao D { D D.. D} D a vecor of arbue values Obecve: r o esmae e uerlg rue probabl srbuo over varables X px usg

More information

Final Exam Applied Econometrics

Final Exam Applied Econometrics Fal Eam Appled Ecoomercs. 0 Sppose we have he followg regresso resl: Depede Varable: SAT Sample: 437 Iclded observaos: 437 Whe heeroskedasc-cosse sadard errors & covarace Varable Coeffce Sd. Error -Sasc

More information

Synchronization of Complex Network System with Time-Varying Delay Via Periodically Intermittent Control

Synchronization of Complex Network System with Time-Varying Delay Via Periodically Intermittent Control Sychrozao of Complex ework Sysem wh me-varyg Delay Va Perodcally Ierme Corol JIAG Ya Deparme of Elecrcal ad Iformao Egeerg Hua Elecrcal College of echology Xaga 4, Cha Absrac he sychrozao corol problem

More information

A Modular On-line Profit Sharing Approach in Multiagent Domains

A Modular On-line Profit Sharing Approach in Multiagent Domains A Modular O-le Prof Sharg Approach Mulage Domas Pucheg Zhou, ad Bgrog Hog Absrac How o coordae he behavors of he ages hrough learg s a challegg problem wh mul-age domas. Because of s complexy, rece work

More information

The Properties of Probability of Normal Chain

The Properties of Probability of Normal Chain I. J. Coep. Mah. Sceces Vol. 8 23 o. 9 433-439 HIKARI Ld www.-hkar.co The Properes of Proaly of Noral Cha L Che School of Maheacs ad Sascs Zheghou Noral Uversy Zheghou Cy Hea Provce 4544 Cha cluu6697@sa.co

More information

Voltage Sensitivity Analysis in MV Distribution Networks

Voltage Sensitivity Analysis in MV Distribution Networks Proceedgs of he 6h WSEAS/IASME I. Cof. o Elecrc Power Sysems, Hgh olages, Elecrc Maches, Teerfe, Spa, December 6-8, 2006 34 olage Sesvy Aalyss M Dsrbuo Neworks S. CONTI, A.M. GRECO, S. RAITI Dparmeo d

More information

EE 6885 Statistical Pattern Recognition

EE 6885 Statistical Pattern Recognition EE 6885 Sascal Paer Recogo Fall 005 Prof. Shh-Fu Chag hp://www.ee.columba.edu/~sfchag Lecure 5 (9//05 4- Readg Model Parameer Esmao ML Esmao, Chap. 3. Mure of Gaussa ad EM Referece Boo, HTF Chap. 8.5 Teboo,

More information

General Complex Fuzzy Transformation Semigroups in Automata

General Complex Fuzzy Transformation Semigroups in Automata Joural of Advaces Compuer Research Quarerly pissn: 345-606x eissn: 345-6078 Sar Brach Islamc Azad Uversy Sar IRIra Vol 7 No May 06 Pages: 7-37 wwwacrausaracr Geeral Complex uzzy Trasformao Semgroups Auomaa

More information

RATIO ESTIMATORS USING CHARACTERISTICS OF POISSON DISTRIBUTION WITH APPLICATION TO EARTHQUAKE DATA

RATIO ESTIMATORS USING CHARACTERISTICS OF POISSON DISTRIBUTION WITH APPLICATION TO EARTHQUAKE DATA The 7 h Ieraoal as of Sascs ad Ecoomcs Prague Sepember 9-0 Absrac RATIO ESTIMATORS USING HARATERISTIS OF POISSON ISTRIBUTION WITH APPLIATION TO EARTHQUAKE ATA Gamze Özel Naural pulaos bolog geecs educao

More information

Regression Approach to Parameter Estimation of an Exponential Software Reliability Model

Regression Approach to Parameter Estimation of an Exponential Software Reliability Model Amerca Joural of Theorecal ad Appled Sascs 06; 5(3): 80-86 hp://www.scecepublshggroup.com/j/ajas do: 0.648/j.ajas.060503. ISSN: 36-8999 (Pr); ISSN: 36-9006 (Ole) Regresso Approach o Parameer Esmao of a

More information

On an algorithm of the dynamic reconstruction of inputs in systems with time-delay

On an algorithm of the dynamic reconstruction of inputs in systems with time-delay Ieraoal Joural of Advaces Appled Maemacs ad Mecacs Volume, Issue 2 : (23) pp. 53-64 Avalable ole a www.jaamm.com IJAAMM ISSN: 2347-2529 O a algorm of e dyamc recosruco of pus sysems w me-delay V. I. Maksmov

More information

Abstract. Keywords: Mutation probability, evolutionary computation, optimization, sensitivity, variability. 1. Introduction. 2. Proposed Algorithm

Abstract. Keywords: Mutation probability, evolutionary computation, optimization, sensitivity, variability. 1. Introduction. 2. Proposed Algorithm EgOp 2008 Ieraoal Coferece o Egeerg Opmzao Ro de Jaero, Brazl, 01-05 Jue 2008. Absrac Redefg Muao Probables for Evoluoary Opmzao Problems Raja Aggarwal Faculy of Egeerg ad Compuer Scece Cocorda Uversy,

More information

Other Topics in Kernel Method Statistical Inference with Reproducing Kernel Hilbert Space

Other Topics in Kernel Method Statistical Inference with Reproducing Kernel Hilbert Space Oher Topcs Kerel Mehod Sascal Iferece wh Reproducg Kerel Hlber Space Kej Fukumzu Isue of Sascal Mahemacs, ROIS Deparme of Sascal Scece, Graduae Uversy for Advaced Sudes Sepember 6, 008 / Sascal Learg Theory

More information

Stabilization of LTI Switched Systems with Input Time Delay. Engineering Letters, 14:2, EL_14_2_14 (Advance online publication: 16 May 2007) Lin Lin

Stabilization of LTI Switched Systems with Input Time Delay. Engineering Letters, 14:2, EL_14_2_14 (Advance online publication: 16 May 2007) Lin Lin Egeerg Leers, 4:2, EL_4_2_4 (Advace ole publcao: 6 May 27) Sablzao of LTI Swched Sysems wh Ipu Tme Delay L L Absrac Ths paper deals wh sablzao of LTI swched sysems wh pu me delay. A descrpo of sysems sablzao

More information

Complementary Tree Paired Domination in Graphs

Complementary Tree Paired Domination in Graphs IOSR Joural of Mahemacs (IOSR-JM) e-issn: 2278-5728, p-issn: 239-765X Volume 2, Issue 6 Ver II (Nov - Dec206), PP 26-3 wwwosrjouralsorg Complemeary Tree Pared Domao Graphs A Meeaksh, J Baskar Babujee 2

More information

An Efficient Dual to Ratio and Product Estimator of Population Variance in Sample Surveys

An Efficient Dual to Ratio and Product Estimator of Population Variance in Sample Surveys "cece as True Here" Joural of Mahemacs ad ascal cece, Volume 06, 78-88 cece gpos Publshg A Effce Dual o Rao ad Produc Esmaor of Populao Varace ample urves ubhash Kumar Yadav Deparme of Mahemacs ad ascs

More information

Complete Identification of Isotropic Configurations of a Caster Wheeled Mobile Robot with Nonredundant/Redundant Actuation

Complete Identification of Isotropic Configurations of a Caster Wheeled Mobile Robot with Nonredundant/Redundant Actuation 486 Ieraoal Joural Sugbok of Corol Km Auomao ad Byugkwo ad Sysems Moo vol 4 o 4 pp 486-494 Augus 006 Complee Idefcao of Isoropc Cofguraos of a Caser Wheeled Moble Robo wh Noreduda/Reduda Acuao Sugbok Km

More information

Model for Optimal Management of the Spare Parts Stock at an Irregular Distribution of Spare Parts

Model for Optimal Management of the Spare Parts Stock at an Irregular Distribution of Spare Parts Joural of Evromeal cece ad Egeerg A 7 (08) 8-45 do:0.765/6-598/08.06.00 D DAVID UBLIHING Model for Opmal Maageme of he pare ars ock a a Irregular Dsrbuo of pare ars veozar Madzhov Fores Research Isue,

More information

EMD Based on Independent Component Analysis and Its Application in Machinery Fault Diagnosis

EMD Based on Independent Component Analysis and Its Application in Machinery Fault Diagnosis 30 JOURNAL OF COMPUTERS, VOL. 6, NO. 7, JULY 0 EMD Based o Idepede Compoe Aalyss ad Is Applcao Machery Faul Dagoss Fegl Wag * College of Mare Egeerg, Dala Marme Uversy, Dala, Cha Emal: wagflsky997@sa.com

More information

CS344: Introduction to Artificial Intelligence

CS344: Introduction to Artificial Intelligence C344: Iroduco o Arfcal Iellgece Puhpa Bhaacharyya CE Dep. IIT Bombay Lecure 3 3 32 33: Forward ad bacward; Baum elch 9 h ad 2 March ad 2 d Aprl 203 Lecure 27 28 29 were o EM; dae 2 h March o 8 h March

More information

Neural Network Global Sliding Mode PID Control for Robot Manipulators

Neural Network Global Sliding Mode PID Control for Robot Manipulators Neural Newor Global Sldg Mode PID Corol for Robo Mapulaors. C. Kuo, Member, IAENG ad Y. J. Huag, Member, IAENG Absrac hs paper preses a eural ewor global PID-sldg mode corol mehod for he racg corol of

More information

Development of Hybrid-Coded EPSO for Optimal Allocation of FACTS Devices in Uncertain Smart Grids

Development of Hybrid-Coded EPSO for Optimal Allocation of FACTS Devices in Uncertain Smart Grids Avalable ole a www.scecedrec.com Proceda Compuer Scece 6 (011) 49 434 Complex Adapve Sysems, Volume 1 Cha H. Dagl, Edor Chef Coferece Orgazed by ssour Uversy of Scece ad Techology 011- Chcago, IL Developme

More information

Modified Integrated Multi-Point Approximation And GA Used In Truss Topology Optimization

Modified Integrated Multi-Point Approximation And GA Used In Truss Topology Optimization Joural of Muldscplary Egeerg Scece ad echology (JMES) Vol. 4 Issue 6, Jue - 2017 Modfed Iegraed Mul-Po Appromao Ad GA sed I russ opology Opmzao Adurahma M. Hasse 1, Mohammed A. Ha 2 Mechacal ad Idusral

More information

A Novel ACO with Average Entropy

A Novel ACO with Average Entropy J. Sofware Egeerg & Applcaos, 2009, 2: 370-374 do:10.4236/jsea.2009.25049 Publshed Ole December 2009 (hp://www.scrp.org/joural/jsea) A Novel ACO wh Average Eropy Yacag LI College of Cvl Egeerg, Hebe Uversy

More information

EE 6885 Statistical Pattern Recognition

EE 6885 Statistical Pattern Recognition EE 6885 Sascal Paer Recogo Fall 005 Prof. Shh-Fu Chag hp://.ee.columba.edu/~sfchag Reve: Fal Exam (//005) Reve-Fal- Fal Exam Dec. 6 h Frday :0-3 pm, Mudd Rm 644 Reve Fal- Chap 5: Lear Dscrma Fucos Reve

More information

ASYMPTOTIC EQUIVALENCE OF NONPARAMETRIC REGRESSION AND WHITE NOISE. BY LAWRENCE D. BROWN 1 AND MARK G. LOW 2 University of Pennsylvania

ASYMPTOTIC EQUIVALENCE OF NONPARAMETRIC REGRESSION AND WHITE NOISE. BY LAWRENCE D. BROWN 1 AND MARK G. LOW 2 University of Pennsylvania The Aals of Sascs 996, Vol., No. 6, 38398 ASYMPTOTIC EQUIVALENCE OF NONPARAMETRIC REGRESSION AND WITE NOISE BY LAWRENCE D. BROWN AND MARK G. LOW Uversy of Pesylvaa The prcpal resul s ha, uder codos, o

More information

Survival Prediction Based on Compound Covariate under Cox Proportional Hazard Models

Survival Prediction Based on Compound Covariate under Cox Proportional Hazard Models Ieraoal Bomerc Coferece 22/8/3, Kobe JAPAN Survval Predco Based o Compoud Covarae uder Co Proporoal Hazard Models PLoS ONE 7. do:.37/oural.poe.47627. hp://d.plos.org/.37/oural.poe.47627 Takesh Emura Graduae

More information

Cyclically Interval Total Colorings of Cycles and Middle Graphs of Cycles

Cyclically Interval Total Colorings of Cycles and Middle Graphs of Cycles Ope Joural of Dsree Mahemas 2017 7 200-217 hp://wwwsrporg/joural/ojdm ISSN Ole: 2161-7643 ISSN Pr: 2161-7635 Cylally Ierval Toal Colorgs of Cyles Mddle Graphs of Cyles Yogqag Zhao 1 Shju Su 2 1 Shool of

More information

The Signal, Variable System, and Transformation: A Personal Perspective

The Signal, Variable System, and Transformation: A Personal Perspective The Sgal Varable Syem ad Traformao: A Peroal Perpecve Sherv Erfa 35 Eex Hall Faculy of Egeerg Oule Of he Talk Iroduco Mahemacal Repreeao of yem Operaor Calculu Traformao Obervao O Laplace Traform SSB A

More information