Exact Dynamic Programming for Decentralized POMDPs with Lossless Policy Compression

Size: px
Start display at page:

Download "Exact Dynamic Programming for Decentralized POMDPs with Lossless Policy Compression"

Transcription

1 Proceedngs of he Egheenh Inernaonal Conference on Auomaed Plannng and Schedulng (ICAPS 2008) Exac Dynamc Programmng for Decenralzed POMDPs wh Lossless Polcy Compresson Abdeslam Boularas and Brahm Chab-draa Compuer Scence & Sofware Engneerng Dep. Laval Unversy, Quebec G1k 7p4, CANADA Absrac Hgh dmensonaly of belef space n DEC-POMDPs s one of he maor causes ha makes he opmal on polcy compuaon nracable. The belef sae for a gven agen s a probably dsrbuon over he sysem saes and he polces of oher agens. Belef compresson s an effcen POMDP approach ha speeds up plannng algorhms by proecng he belef sae space o a low-dmensonal one. In hs paper, we nroduce a new mehod for solvng DEC-POMDP problems, based on he compresson of he polcy belef space. The reduced polcy space conans sequences of acons and observaons ha are lnearly ndependen. We esed our approach on wo benchmark problems, and he prelmnary resuls confrm ha Dynamc Programmng algorhm scales up beer when he polcy belef s compressed. Inroducon Decson makng under sae uncerany s one of he greaes challenges n arfcal nellgence. Sae uncerany s a drec resul of sochasc acons and nosed, or alased, observaons. Parally Observable Markov Decson Processes (POMDPs) provde a powerful Bayesan model o solve hs problem (Smallwood & Sondk 1971). In hs model, he sae s represened by a probably dsrbuon over all he possble saes, ha we call a belef sae. The complexy of POMDPs algorhms, whch s proved o be PSPACEcomplee (Papadmrou & Tsskls 1987), depends heavly on he dmenson of he belef sae. Durng he las wo decades, sgnfcan effors have been devoed o developng fas algorhms for large POMDPs, and nowadays, even problems wh a housand of saes can be solved whn a few seconds (Vrn e al. 2007). The rse of applcaons requrng cooperaon beween dfferen agens, lke roboc eams, dsrbued sensors and communcaon neworks, has made he presence of many decson makers a key challenge for buldng auonomous agens. For hs purpose, a generalzaon of POMDPs o Copyrgh c 2008, Assocaon for he Advancemen of Arfcal Inellgence ( All rghs reserved. mul-agen domans, called DEC-POMDPs (Decenralzed POMPDs), was nroduced n (Bernsen, Immerman, & Zlbersen 2002), and snce hen, hs framework has been recevng a growng amoun of aenon. Ths research s bascally movaed by he fac ha many real world problems need o be formalzed as DEC-POMDPs, whle plannng wh DEC-POMDPs s NEXP-complee (Bernsen, Immerman, & Zlbersen 2002), and even fndng ε-opmal soluons s NEXP-hard (Rabnovch, Goldman, & Rosenschen 2003). Fndng good soluons o DEC-POMDPs s so dffcul because here s no opmaly creron for he polces of a sngle agen alone: wheher a gven polcy s beer or worse han anoher depends on he polces of he oher agens. Consequenly, he dmensonaly of he polcy space s a crucal facor n he scalably of DEC-POMDPs algorhms. In hs paper, we propose a new mehod for scalng up Dynamc Programmng for DEC-POMDPs, based on a lossless compresson he polcy space. Our approach s based on he followng observaon: gven a se of polces, only a few sequences are necessary o represen all he polces. Ths mehod s an adapaon o DEC-POMDPs of anoher approach ha was orgnally proposed o reduce he sae space dmensonaly n POMDPs, and whch s known as he Predcve Sae Represenaons (PSRs) (Lman, Suon, & Sngh 2001). In PSRs, saes are replaced by sequences of acons and observaons ha have lnearly ndependen probables. Smlarly, a poll of polces can be represened by a smaller se of sequences ha have lnearly ndependen probables of occurrng n hese polces. Relaed Work A brue force soluon for DEC-POMDPs consss n performng an exhausve search n he space of on polces (Bernsen, Immerman, & Zlbersen 2002), bu hs approach s almos useless n pracce, even for he smalles domans. In fac, he search should focus only on he polces ha are lkely o be domnan. There are wo man approaches for fndng hese polces: op-down heurscs and boom-up dynamc programmng. MAA* was he frs 20

2 algorhm o use a op-down heursc (Szer, Charplle, & Zlbersen 2005). I s an adapaon of A* usng a heursc evaluaon funcon o prune domnaed nodes, where he nodes correspond o on polces. A bg dsadvanage of such op-down search approaches s ha he sarng pon should be known n advance. On he oher hand, Dynamc Programmng (DP) algorhm, proposed n (Hansen, Bernsen, & Zlbersen 2004), consss n consrucng he opmal polces from leaves o roo by usng eraed polcy elmnaon. DP algorhm can solve problems ha are unfeasble wh he exhausve search, bu keeps all he domnan polces for every pon n he belef space, even hose ha wll never be reached n pracce. Ths problem has been effcenly addressed wh Pon Based Dynamc Programmng (PBDP) algorhm proposed n (Szer & Charplle 2006). PBDP makes use of op-down heursc search o deermne whch belef pons wll be reached durng he execuon, and consrucs he bes polcy from leaves o roo wh DP, by keepng only he polces ha are domnan n he reachable pons. In he same ven, Memory Bounded Dynamc Programmng (MBDP) s an algorhm ha has been proposed recenly n (Seuken & Zlbersen 2007) and whch s close o PBDP, s based on boundng he maxmum number of polces kep n memory afer each eraon. All hese algorhms are based on reducng he number of polces o be evaluaed or kep n memory. Alernavely, we can preserve he orgnal polces space, and use a more compac represenaon of he polces. In decson rees, polces are consruced by combnng mulple sequences of acons and observaons, and he same sequences can be replcaed n dfferen polces. The sequenal represenaon akes advanage of hs characersc: all he possble sequences of a gven lengh are represened explcly, and each polcy s represened by a bnary vecor ha ndcaes whch sequences are conaned n hs polcy. Ths mehod has been appled effcenly o DEC-POMDPs n (Aras, Duech, & Charplle 2007), he proposed algorhm uses a mxed neger lnear program o fnd he opmal polces, where each varable corresponds o a sequence. However, here s no guaranee ha he number of sequences wll no exceed he number of polces, hs can happen when we have a few polces wh large horzons. Decenralzed POMDPs DEC-POMDPs, nroduced n (Bernsen, Immerman, & Zlbersen 2002), are a sragh generalzaon of POMDPs o mul-agen sysems. Formally, a DEC-POMDP wh n agens s a uple I,S,{A },P,{Ω},O,R,T,γ, where: I s a fne se of agens, ndexed 1...n. S s a fne se of saes. A s a fne se of ndvdual acons for agen. A = I A s he se of on acons, and a = a 1,...,a n denoes a on acon. P s a ranson funcon, P(s s, a) s he probably ha he sysem changes from sae s o sae s, when he agens execue he on acon a. Ω s a fne se of ndvdual observaons for agen. Ω = I Ω s he se of on observaons, and o = o 1,...,o n denoes a on observaon. O s an observaon funcon, O( o s, a) s he probably ha he agens observe o when he curren sae s s and he on acon ha led o hs sae was a. R s a reward funcon, where R(s, a) denoes he reward (or cos) gven for execung acon a n sae s. T s he horzon of plannng (he oal number of seps). γ s a dscoun facor. Plannng algorhms for DEC-POMDPs am o fnd he bes on polcy of horzon T, whch s a collecon of several local polces, one for each agen. A local polcy of horzon for agen, denoed by q, s a mappng from local hsores of observaons o 1 o2...o o acons n A. Usually, we use decson rees o represen local polces, each node of he decson ree s labeled wh an ndvdual acon a, and each arc s labeled wh an ndvdual observaon o. To ease noaons, we consder n he remander of hs paper ha we have only wo agens, and, all he resuls can be easly exended o he general case. A on polcy of horzon for agens and s denoed by q = q,q. We also use Q, Q o ndcae he ses of local polces for agens and respecvely, and Q for he se of on polces. Snce he saes are parally observable, agens should choose her acons accordng o a belef on he curren sae. The belef sae, denoed by he vecor b, s a probably dsrbuon over he dfferen saes. In he sngle-agen conex, he belef sae s a suffcen sasc for makng opmal decsons, bu n mul-agen sysems, he reward gven o an ndvdual acon depends on all he acons aken by he oher agens a he same me; he belef sae s hen no longer suffcen for makng opmal decsons. In order o choose opmally s acon, an agen should ake no accoun whch acons he oher agens are gong o execue. Ideally, every agen should know exacly he acons of he oher agens, bu unforunaely, hs canno be acheved whou use of communcaon. In fac, he on polcy s provded o all he agens, and he frs on acon can easly be predced, snce we us have o look a he frs node of every local polcy ree. Bu hen, he nex acons depend on he local observaons perceved by each agen, and whou communcaon, we can only consder a probably dsrbuon over he acons (or he remanng subrees) of he oher agens. Ths dsrbuon s wha we call a mul-agen belef sae. The belef sae b for agen conans a probably dsrbuon over he saes S, and anoher probably dsrbuon over he curren polces Q of agen. Noce ha b (s,q ) s he probably ha he sysem s n sae s and he curren polcy of agen s q. 21

3 Inpu: Q 1, Q 1 and V 1 ; Q, Q fullbackup(q 1 ), fullbackup(q 1 ); Calculae he value vecors V by usng V 1 (equaon 1); repea remove he polces of Q ha are domnaed (Table 1); remove he polces of Q ha are domnaed (Table 1); unl no more polces n Q or Q can be removed ; Oupu: Q,Q and V ; Algorhm 1: Dynamc Programmng for DEC- POMDPs (Hansen, Bernsen, & Zlbersen 2004). Dynamc Programmng for DEC-POMDPs Dynamc Programmng s by far he echnque mos used for solvng mulsage decson problems, where he opmal polces of horzon are recursvely consruced from he opmal sub-polces of horzon 1. Ths mehod has been wdely used for fndng opmal fne horzon polces for POMDPs snce (Smallwood & Sondk 1971) presened he value eraon algorhm. Recenly, (Hansen, Bernsen, & Zlbersen 2004) proposed an neresng exenson of he value eraon algorhm o decenralzed POMDPs, called Dynamc Programmng Operaor for DEC-POMDPs. We revew here brefly he prncpal seps of hs algorhm. The expeced dscouned reward of a on polcy q, sared from sae s, s gven recursvely by Bellman value funcon: V q (s) = R(s, A(q ))+γ P(s s, A(q )) O( o s, A(q ))V o(q )(s ) s S o Ω (1) where A(q ) s he frs on acon of he polcy q (he roo node), o s a on observaon, and o(q ) s he sub-polcy of q below he roo node and he observaon o. The value of an ndvdual polcy q, accordng o a belef sae b, s gven by he followng funcon: V q (b ) = b (s,q )V q s S q,q (s) (2) Q where q,q denoes he on polcy made up of q and q, V q,q (s) s gven by equaon 1. The Dynamc Programmng Operaor (Algorhm 1) fnds he opmal polces of horzon, gven he opmal polces of horzon ( 1). V s he se of value vecors V q correspondng o he on polces of horzon. Frs, he ses Q, Q are generaed by exendng he polces of Q 1, Q 1, and V are calculaed by usng V 1 n equaon 1, hen he weakly domnaed polces of each agen are eravely prune. The prunng process sops when no more polces can be removed from Q or Q. A polcy q s sad o be weakly domnaed f and only f: b (S Q ), q Q {q }: V q (b ) V q (b ) (3) mnmze: ε subec o: s S s S, q Q : s S b (s,q ) = 1 q Q 0 b (s,q ) 1 q Q {q } : b (s,q )[V q q Q,q (s) V q,q (s)] + ε > 0 Table 1: The lnear program used o check f a polcy q s domnaed or no (Hansen, Bernsen, & Zlbersen 2004). Movaon Polcy Space Compresson The man problem wh DEC-POMDPs, compared o POMDPs, s he dmensonaly of he polcy space. In fac, he number of polces grows double exponenally wh respec o he plannng horzon and he number of observaons. If we ge Q 1 polces for agen a sep 1, hen A Q 1 Ω new polces wll be creaed a sep. Ths curse of dmensonaly has dramac consequences on boh me and space complexy of he DEC-POMDPs algorhms. From Algorhm 1, we can see ha he dynamc programmng operaor spends mos of s me deermnng he weakly domnaed polces by checkng he nequaly (3) for every polcy q. The usual approach for performng hs es s o use he lnear program of Table 1. The obecve funcon o be mnmzed s defned by ε, whch s he greaes dfference beween he value of q and he value of any oher polcy q over he belef space. If ε 0 hen he polcy q s domnaed and should be removed, and f ε < 0, hen here s some regon n (S Q ) where q s domnan. The varables are ε and he probables b(.,.) of he mul-agen belef sae, so here are S Q + 1 varables. The me complexy of a lnear program solver depends on he number of varables and consrans defned n he problem, so, depends drecly on he number of polces and he way we represen he belefs over hese polces. However, he man problem of Dynamc Programmng (Algorhm 1) remans he sze of he memory space requred o represen he value vecors V for each on polcy. The memory space requred o represen hese vecors a sep s Q Q S floas. Indeed, hs algorhm runs ou of memory several eraons before runnng ou of me. 22

4 a 2 a 1 o 1 o 1 a 1 Q = {q a,q b,q c,q d } o 2 o 2 a 1 a 2 a 1 o 1 o 2 o 1 o 2 a 1 a 1 a 1 a 3 o 1 o 2 a 1 a 1 a 2 o 1 o 2 a 2 a 1 q a a 1 o 1 o 2 a 1 a 3 o 1 o 2 o 1 o 2 a 1 a 1 a 2 a 1 o 1 o 2 o 1 o 2 a 2 a 1 a 2 a 3 q b o 1 o 2 a 2 a 3 { Bass( Q ) }} { q 1 q 2 q 3 U = a 1 o 1 a 2 o 1 a 2 a 1 o 1 a 2 o 1 a 1 = q a q b q c q d a 1 o 1 a 2 o 2 a 3 a 1 o 1 a 2 o 2 a a 1 o 2 a 1 o 2 a 3 a 1 o 2 a 1 o 2 a 1 a 1 o 2 a 1 o 1 a 2 a 1 o 2 a 1 o 1 a 1 dmenson = 4 ( ) dmenson = 3 q c q d F Ũ Fgure 1: Reducng he polcy space dmensonaly. To allevae hese problems, we need o use a more compac echnque o represen he belef of agen on he polces of agen, nsead of he nave probably dsrbuon over all he polces Q. We can explo he srucure of he polces, and fnd some feaures Q = { q 1,..., q Q } whch consue a suffcen nformaon: each pon n Q wll correspond o a pon n Q, and vce versa. We should also guaranee ha Q Q and ha mos of he me Q Q. A polcy s a collecon of sequences of acons and observaons. A sequence q of horzon for agen s an ordered ls of ndvdual acons and 1 ndvdual observaons (a 1,o1,...,o 1,a ). A on sequence q of horzon s a couple q, q where q s an ndvdual sequence for agen and q s an ndvdual sequence for agen, a on sequence can also be seen as one ordered ls of on acons and observaons. The se Q can be compleely replaced, whou any loss of nformaon, by a marx U called he oucome marx (Sngh, James, & Rudary 2004). Each row of U wll correspond o a polcy q Q and each column wll correspond o a sequence q. U (q, q ) s defned as he probably ha agen wll execue he acons of q f he observaons of q occur, such ha he acual polcy of agen s q. Snce he polces are deermnsc, we have U (q, q ) = 1 f he sequence q appears n he polcy q, and U (q, q ) = 0 else. Fgure 1 shows a se Q conanng 4 ndvdual polces for agen : q a,q b,q c and q d. There are 8 dfferen sequences n hese polces, he marx U has hen 4 rows and 8 columns. If b (s,.) s a mul-agen belef sae,.e. a probably dsrbuon over he polces of Q for some sae s S, hen he produc b (s,.)u reurns a vecor conanng he probably of every sequence of Q. To reduce he dmensonaly of b (s,.) from Q o N, we should fnd a ransformaon funcon f defned by: f : Q [0,1] N b (s,.) b (s,.) b (s,.) s called he reduced mul-agen belef sae, corresponds o he belef abou he sequences, whereas b (s,.) s he belef abou he polces. In order ha f be an accurae ransformaon funcon, we should be able o make predcons abou any sequence q by usng only he vecor b (s,.). Lnear reducon of polcy space dmensonaly The oucome marx U can be facorzed as follows: U = F Ũ In hs case, our ransformaon funcon s smply he marx F. The reduced belef sae b can be generaed from b by: b (s,.) = b (s,.)f and he probables of Q sequences can be found by: b (s,.)u = b (s,.)(f Ũ ) = b (s,.)ũ The marx F s a bass for he marx U, s formed by lnearly ndependen columns of U. We use Bass( Q ) o ndcae he se of sequences correspondng o hese lnearly ndependen columns. In Fgure 1, Bass( Q ) conans he sequences q 1, q 2 and q 3. Noe ha Bass( Q ) Q, because he lnear rank of a marx canno be more han he number of rows n hs marx, and F has exacly Q rows. Snce F s a bass for he oucome marx U, we can wre any column of U as lnear combnaon of F columns: Pr( q q ) = U (q, q ) = F (q,.)w q 23

5 where w q s a wegh vecor 1 assocaed o he sequence q (we have one wegh vecor per sequence). Smlarly, w q s he column correspondng o q n he marx Ũ. If b (s,.) = b (s,.)f s a reduced belef sae, hen: Pr(s, q b ) = b (s,q )Pr( q q ) q Q = b (s,q )(F(q,.)w q ) q Q = b (s,.)w q Ths means ha gven a reduced belef sae, he probably of any sequence s a lnear combnaon of he probables conaned n hs reduced belef sae. Fndng he bass sequences The man problem now s how o fnd he bass sequences of Q and her marx F whou facorzng he enre marx U a each sep. The heorem below saes ha f Bass( Q 1 ) s he se of bass sequences for horzon 1, hen he sequences of Bass( Q ) for horzon wll be among he one sep exensons of he Bass( Q 1 ) sequences. Ths means ha we need o exend only he sequences of Bass( Q 1 ) a each sep, and consruc he marx F where he columns correspond o hese exended sequences. F may conan some lnearly dependen columns, along wh he bass columns. Thus, we use a Gauss-Jordan elmnaon on F, or any oher decomposon echnque, o exrac Bass( Q ) n a polynomal me. Theorem 1. Le Q 1 be a se of horzon 1 polces, Q 1 s he se of horzon 1 sequences correspondng o Q 1, Q s he se of horzon polces creaed from Q 1 by an exhausve backup, and Q s he se of horzon sequences correspondng o Q, hen: Bass( Q ) {ao q 1 : a A,o Ω, q 1 Bass( Q 1 )} U 1 of U 1 s he oucome marx of Q 1, F 1 s he sub-marx correspondng o he sequences of Bass( Q 1 ). Le q be a polcy of Q, q 1 a sequence of Q 1, a an acon of A and o an observaons of Ω, hen we can be n one of he followng wo suaons: Case 1: he polcy ree q sars wh an acon dfferen from a, hen he sequence ao q 1 does no appear n he polcy q, so: U (q,ao q 1 ) = 0. Case 2: he polcy ree q sars wh he acon a, hen he sequence ao q 1 appears n he polcy q ff q 1 appears n o(q ), he sub-ree of q below he frs acon a and he observaon o. We have hen U (q,ao q 1 ) = U 1 (o(q ), q 1 ) and q Bass( Q 1 ) : F (q,ao q ) = F 1 (o(q ), q ). So 2 : U (q,ao q 1 ) = U 1 where we defne w ao q 1 { wao q 1 w ao q 1 (o(q = q Bass( Q 1 ) ), q 1 ) F 1 (o(q ), q )w q 1 ( q ) = F (q,ao q )w ao q 1 (ao q ) q Bass( Q 1 ) = F (q,.)w ao q 1 as follows: (a o q ) = w q 1 ( q ) f a = a and o = o, (a o q ) = 0 else. Cases 1 and 2 can be grouped ogeher, we have hen: U (q,ao q 1 ) = F (q,.)w ao q 1 Noce ha he vecor w ao q 1 s defned ndependenly on he polcy q. So, he sequences {ao q : a A,o Ω, q Bass( Q 1 )} are ndeed a bass for he marx U. Proof. Ths heorem clams ha he sequences of Bass( Q ) are conaned n he one sep exensons of he Bass( Q 1 ) sequences. In oher words, q 1 Q 1, a A, o Ω, we should be able o wre he column U (.,ao q ) as a lnear, whch s he sub-marx combnaon of he columns of F of U correspondng o he sequences ao q 1 Ω, q 1 : a A,o Bass( Q 1 ). U s he oucome marx of Q, 1 To smplfy he noaons, we consder ha he belef vecors are rows and he wegh vecors are columns. Reduced value vecors Snce we wan o use hs represenaon for plannng, we have o redefne he value funcon (equaons 1 and 2), gven ha he belef sae s now abou sequences nsead of polces. Frs, we defne he expeced value V q (s) of a on sequence q = a 1 o 1 a 2 o 2... a n sae s as follows: V q (s) = Pr( q s)r q (s) = Pr( o 1 o 2... o 1 s 1 = s, a 1 a 2... a 1 )R q (s) 2 We use q o ndcae a bass sequence, he horzon of hs sequence can be nferred from he conex. 24

6 Pr( o 1 o 2... o 1 s, a 1 a 2... a 1 ) s he probably ha he observaons of q wll occur f we sar execung he acons of q a s, and R q (s) s he reward expeced from he acons of q such ha he observaons of q wll occur. R q (s) s gven by: R q (s) = γ k 1 Pr(s k = s s 1 = s, a 1 o 1... a )R(s, a k ) k=1 s S = k=1 γk 1 s S Pr(s k = s, o 1... o 1 s 1 = s, a 1... a 1 )R(s, a k ) Pr( o 1 o 2... o 1 s 1 = s, a 1 a 2... a 1 ) = k=1 γk 1 s S α k (s,s, q )β k (s, q )R(s, a k ) Pr( o 1 o 2... o 1 s 1 = s, a 1 a 2... a 1 ) where: { αk (s,s, q ) = Pr( o 1... o k 1,s k = s s 1 = s, a 1... a k 1 ) β k (s, q ) = Pr( o k... o 1 s k = s, a k a 2... a 1 ) We appled Bayes rule, and hen we decomposed Pr(s k = s, o 1... o 1 s 1 = s, a 1... a 1 ) no α k (s,s, q ) and β k (s, q ), akng advanage of Markov propery. So, he expeced value of a on sequence q s gven by: V q (s) = γ k 1 α k (s,s, q )β k (s, q )R(s, a k ) k=1 s S Ths value s calculaed recursvely as follows: V a o q 1(s) = γ k 1 α k (s,s, a o q 1 )β k (s, a o q 1 )R(s, a) k=1 s S = β 1 (s, a o q 1 )R(s, a) + γ k 1 α k (s,s, a o q 1 )β k (s, a o q 1 )R(s, a) k=2 s S = β 1 (s, a o q 1 )R(s, a) 1 +γ γ k 1 [P(s s, a)p( o s, a)α k (s,s, q 1 ) k=1 s S s S β k (s, q 1 )R(s, a)] = β 1 (s, a o q 1 )R(s, a)+γ P(s s, a)p( o s, a)v q 1(s ) s S We used he followng properes: α 1 (s,s, a o q 1 ) = 1, α 1 (s,s, a o q 1 ) = 0 for s s, β k (s, a o q 1 ) = β k 1 (s, q 1 ) for k 2, and α k (s,s, a o q 1 ) = s S P(s s, a)p( o s, a)α k 1 (s,s, q 1 ). The value funcon of an ndvdual polcy q n a reduced belef sae b s gven by: V q ( b ) = Pr(s b ) Pr( q b )Pr( q q )V q s S q Q, q (s) q Q = s S q Bass( Q ) q Bass( Q ) F (q, q )=1 b (s, q )Ṽ q, q (s) where: Ṽ q, q (s) de = f w q ( q )w q ( q q Q )V q, q (s) q Q We subsued he probably of each sequence q wh a lnear combnaon of he probables b (s, q ) of he bass sequences q, and he probably of each sequence q wh a lnear combnaon of he probables F (q, q ) {0,1} of he bass sequences q. The reduced vecor Ṽ q, q defnes he conrbuon of q, q o he value of a on polcy, by ncludng a proporon of he values of all he sequences whch depend on q, q. If we wan o calculae he value of a polcy q for a gven mul-agen belef sae b, all we need are he vecors Ṽ q, q and he marx F. We wll see now how o calculae he value vecors Ṽ gven he value vecors Ṽ 1. A horzon 1, each polcy s a sngle acon. We have hen Bass(Q 1 ) = A, and Ṽ a,a (s) = V a,a (s) = R(s, a,a ). A horzon > 1, we know from Theorem 1 ha he bass sequences are of he form a o q and a o q, where q Bass( Q 1 ) and q Bass( Q 1 ). Ṽ a o q,a o q (s) = where: C a o q,a o q (s) q 1 Q 1 q 1 Q 1 w a o q 1 ( q )w a o q 1 ( q ) V a o q 1,a o q 1 (s) = R(s, a,a )C a o q,a o q (s) + γ Pr( o,o,s s, a,a )V q, q (s ) s S (4) de f = q 1 Q 1 q 1 Q 1 w a o q 1 Pr( a o q 1 ( q )w a o q 1 ( q ),a o q 1 ) s) = P(s s, a,a )O( o,o s, a,a )C q, q s (s ) (5) S In order o calculae he value vecor Ṽ a o q,a o q, we only need o know he vecors C q, q and Ṽ q, q, provded by he las eraon of he dynamc programmng algorhm. Algorhm Algorhm 2 descrbes he man seps of he dynamc programmng algorhm where he polces are evaluaed n a reduced dmensonal space. We keep he same srucures Q and Q used n Algorhm 1. The value vecors V are replaced by lower dmensonal vecors Ṽ. We also need o 25

7 Inpu: Q 1,Q 1,Bass( Q 1 ),Bass( Q 1 ), Ṽ 1,Ṽ 1,C 1,C 1 ; Q, Q fullbackup(q 1 ), fullbackup(q 1 ); Bass( Q ) A O Bass( Q 1 ); Bass( Q ) A O Bass( Q 1 ); Calculae he vecors C by usng C 1 (Equaon 5); Calculae he vecors Ṽ by usng Ṽ 1 (Equaon 4); repea remove he polces of Q ha are domnaed (Table 2); remove he polces of Q ha are domnaed (Table 2); unl no more polces n Q or Q can be removed ; removedependence(q,bass( Q ),Bass( Q ),C,Ṽ ); removedependence(q,bass( Q ),Bass( Q ),C,Ṽ ); Oupu: Q, Q, Bass( Q ), Bass( Q ), Ṽ, Ṽ, C ; Algorhm 2: Dynamc Programmng for DEC- POMDPs wh Lossless Polcy Belef Compresson. Inpu: Q,Bass( Q ),Bass( Q ),C,Ṽ ; Use a decomposon mehod o fnd he lnearly dependen sequences n Bass( Q ), and remove hem from Bass( Q ); foreach each removed sequence q from Bass( Q ) do foreach bass sequence q from Bass( Q ) do foreach bass sequence q from Bass( Q ) do C q, q C q, q + w q ( q )C q, q ; Ṽ q, q Ṽ q, q + w q ( q )Ṽ q, q ; end end end Oupu: Updaed Bass( Q ),C,Ṽ ; Algorhm 3: Removng he dependen sequences from Bass( Q ) and updang he vecors C, Ṽ. manan he lss Bass( Q ), Bass( Q ), and he probably vecors C. The marx F (resp. F ) s mplcly represened by specfyng for each bass sequence he ls of polces where hs sequence occurs. A sep = 1, we have Q 1 = Bass( Q 1 ) = A,Q 1 = Bass( Q 1 ) = A and a A, a A : Ṽ a,a = R(., a,a ),C a,a = 1. A sep > 1, Q and Q are he ses of all possble polces of horzon where he subpolces of horzon 1 are n Q 1 and Q 1 respecvely. Bass( Q ) and Bass( Q ) are formed by one sep exensons of Bass( Q 1 ) and Bass( Q 1 ) respecvely. We now calculae he probably vecors C by usng he vecors C 1 n Equaon (5), and he value vecors Ṽ usng he vecors C and Ṽ 1 n Equaon (4). The vecors Ṽ are used o deermne whch polces of and are domnaed and should be removed. We use he lnear program of Table 2 o solve hs problem for each polcy q. The varables of hs lnear program are: ε, and he probablmnmze: ε subec o: s S, q Bass( Q ) : q Q {q } : 0 b (s, q ) 1 V q ( b ) V q ( b ) + ε > 0 Table 2: The lnear program used o deermne f a polcy q s domnaed or no, wh a reduced belef space. es b(.,.) of he reduced mul-agen belef sae, we have hen S Bass( Q ) +1 varables. Noce ha conrary o he orgnal belef sae b (s,.) whch s a probably dsrbuon over he polces of agen, he reduced belef space b (s,.) s no necessarly a probably dsrbuon, because he sequences are no muually exclusve. In Fgure 1 for example, f b(s,q a ) = 1 hen b(s, q 1 ) = 1 and b(s, q 3 ) = 1. The relaon beween he dfferen bass sequences s more complex and more consrans should be added o make sure ha any reduced belef consdered n he lnear program wll really correspond o some belef n he orgnal space. In fac, f we ake any belef b (s,.), we can always fnd a reduced belef b (s,.) = b (s,.)f where all he polces keep he same values, bu gven a reduced belef b (s,.), we are no sure of fndng a belef b (s,.) n he orgnal space such ha b (s,.) = b (s,.)f. Ths s rue f and only f he ransformaon funcon, represened by he marx F s a becon. However, f a polcy q s no domnaed, hen here s a belef b (.,.) where V q (b ) > V q (b ), q Q {q }, and n he correspondng reduced belef b (.,.) = b (.,.)F, we wll also have V q ( b ) > V q ( b ), q Q {q }. Therefore, he lnear program of Table 2. keeps a leas all he domnan polces, bu can keep some domnaed polces (Table 3.). Afer prunng he domnaed polces, some bass sequences become lnearly dependen. Algorhm 3 s hen used o remove he newly dependen sequences, and o updae he parameers C and Ṽ of he remanng bass sequences. These wo updae operaons are derved from he defnons of C and Ṽ. Noce ha when we elmnae polces (.e. elmnae rows from he marx U), he lnearly dependen sequences reman dependen, and keep he same wegh vecors. Expermens We mplemened boh of Algorhm 1 (DP) and Algorhm 2 (DP wh Polcy Compresson) usng ILOG Cplex 10 solver on an AMD Ahlon machne wh a 1.80 GH processor and 1.5 GB ram. We used Gauss-Jordan elmnaon mehod o 26

8 Dynamc Programmng Dynamc Programmng wh Lossless Polcy Compresson Problem T runme polces runme polces bass sequences compresson rao MA-Tger (27,27) 0.17 (27,27) (18,18) (675,675) 1.79 (675,675) (90,90) (195075,195075) (540,540) MABC (8,8) 0.14 (8,8) (8,8) (72,72) 0.36 (72,72) (24,24) (1800,1458) 4.59 (3528,3528) (80,80) 44.1 Table 3: The runme (n seconds) and he number of polces and sequences of DP algorhms, wh and whou compresson. fnd he bass sequences. For he las sep of plannng, we om prunng he domnaed polces snce we wll no use hem o generae furher polces, hus, we only generae he value vecors of each on polcy and each on sequence. We compared he performances of hese wo algorhms on wo benchmark problems MA-Tger and MABC (Hansen, Bernsen, & Zlbersen 2004). Boh of he wo algorhms fnd he same opmal values. However, he memory space used o represen value vecors s sgnfcanly smaller when we use he compresson approach. In fac, he value vecors n he orgnal DP algorhm are defned on saes and polces, whereas he reduced value vecors are defned on saes and bass sequences. Noce also ha he compresson rao (polces number/bass sequences number) grows exponenally w.r. he plannng horzon. Also, he runme of DP s mproved when he compresson algorhm s used. Indeed, he backup of reduced value vecors (equaons 4 and 5) akes less me han he backup of orgnal value vecors (equaon 1). However, gven ha he lnear program of Table 2 s under-consraned, our algorhm can keep some addonal polces ha are domnaed. In MABC for example, our algorhm generaes on polces a he begnnng of horzon 4, whle only on polces are no domnaed. Ths problem can be solved by addng a larger sysem of consrans on he probables of he sequences, bu he compuaonal effcency wll be possbly affeced. We can see ha as n all he compresson echnques, here s a radeoff beween he space performance and he me performance. Concluson The dmensonaly of he polcy space s a crucal facor n he complexy of Dynamc Programmng algorhms for DEC-POMDPs. In hs paper, we nroduced a new approach for dealng wh hs problem, based on proecng he polcy belefs from he hgh dmensonal space of rees o he low dmensonal space of sequences, and usng marx facorzaon mehods o reduce even more he number of sequences. Consequenly, he memory space used n hs algorhm s sgnfcanly smaller, whle he runme s lower compared o he orgnal DP algorhm. Ths mehod can be used n approxmae DP algorhms, manly for domans wh large observaons space. We arge also o nvesgae quck and lossy facorzaon echnques, and more specfcally, bnary-marces facorzaon algorhms. References Aras, R.; Duech, A.; and Charplle, F Mxed Ineger Lnear Programmng for Exac Fne-Horzon Plannng n Decenralzed POMDPs. In Proceedngs of Inernaonal Conference on Auomaed Plannng and Schedulng (ICAPS 07), Bernsen, D.; Immerman, N.; and Zlbersen, S The Complexy of Decenralzed Conrol of Markov Decson Processes. Mahemacs of Operaons Research 27(4): Hansen, E.; Bernsen, D.; and Zlbersen, S Dynamc Programmng for Parally Observable Sochasc Games. In Proceedngs of he 19h Naonal Conference on Arfcal Inellgence (AAAI 04), Lman, M.; Suon, R.; and Sngh, S Predcve Represenaons of Sae. In Advances n Neural Informaon Processng Sysems 14 (NIPS 01), Papadmrou, C., and Tsskls, J The Complexy of Markov Decson Process. Mahemacs of Operaons Research 12(3): Rabnovch, Z.; Goldman, C.; and Rosenschen, J The Complexy of Mulagen Sysems: he Prce of Slence. In Proceedngs of he second nernaonal on conference on Auonomous agens and mulagen sysems (AAMAS 03), Seuken, S., and Zlbersen, S Improved Memory-Bounded Dynamc Programmng for Decenralzed POMDPs. In Proceedngs of he 23rd Conference on Uncerany n Arfcal Inellgence (UAI 07). Sngh, S.; James, M.; and Rudary, M Predcve Sae Represenaons: A New Theory for Modelng Dynamcal Sysems. In Uncerany n Arfcal Inellgence: Proceedngs of he Tweneh Conference (UAI 04). Smallwood, R. D., and Sondk, E. J The Opmal Conrol of Parally Observable Markov Decson Processes over a Fne Horzon. Operaons Research 21(5): Szer, D., and Charplle, F Pon-Based Dynamc Programmng for DEC-POMDPs. In Proceedngs of he 21h Naonal Conference on Arfcal Inellgence (AAAI 06), Szer, D.; Charplle, F.; and Zlbersen, S MAA*: A Heursc Search Algorhm for Solvng Decenralzed POMDPs. In Proceedngs of 21s Conference on Uncerany n Arfcal Inellgence (UAI 05). Vrn, Y.; Shan, G.; Shmony, S.; and Brafman, R Scalng Up: Solvng POMDPs hrough Value Based Cluserng. In Proceedngs of he Tweny-Second AAAI Conference on Arfcal Inellgence (AAAI 07),

9

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

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

More information

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

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

More information

On One Analytic Method of. Constructing Program Controls

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

More information

Variants of Pegasos. December 11, 2009

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

More information

FTCS Solution to the Heat Equation

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

More information

Robustness Experiments with Two Variance Components

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

More information

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

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

More information

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

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

More information

Cubic Bezier Homotopy Function for Solving Exponential Equations

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

More information

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

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

More information

Solution in semi infinite diffusion couples (error function analysis)

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

More information

Dual Approximate Dynamic Programming for Large Scale Hydro Valleys

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

More information

Fall 2010 Graduate Course on Dynamic Learning

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

More information

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

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

More information

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

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

More information

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

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

More information

An introduction to Support Vector Machine

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

More information

Lecture 6: Learning for Control (Generalised Linear Regression)

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

More information

CS286.2 Lecture 14: Quantum de Finetti Theorems II

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

More information

Clustering (Bishop ch 9)

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

More information

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

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

More information

Advanced Macroeconomics II: Exchange economy

Advanced Macroeconomics II: Exchange economy Advanced Macroeconomcs II: Exchange economy Krzyszof Makarsk 1 Smple deermnsc dynamc model. 1.1 Inroducon Inroducon Smple deermnsc dynamc model. Defnons of equlbrum: Arrow-Debreu Sequenal Recursve Equvalence

More information

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

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

More information

Robust and Accurate Cancer Classification with Gene Expression Profiling

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

More information

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

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

More information

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

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

More information

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

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

More information

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

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

More information

Lecture VI Regression

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

More information

Chapter 6: AC Circuits

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

More information

Lecture 11 SVM cont

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

More information

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

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

More information

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

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

More information

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

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

More information

Linear Response Theory: The connection between QFT and experiments

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

More information

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

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

More information

Dynamic Team Decision Theory

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

More information

Graduate Macroeconomics 2 Problem set 5. - Solutions

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

More information

Mechanics Physics 151

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

More information

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

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

More information

CHAPTER 10: LINEAR DISCRIMINATION

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

More information

Comb Filters. Comb Filters

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

More information

Volatility Interpolation

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

More information

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

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

More information

Math 128b Project. Jude Yuen

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

More information

Mechanics Physics 151

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

More information

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

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

More information

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

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

More information

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

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

More information

Mechanics Physics 151

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

More information

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

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

More information

arxiv: v1 [cs.sy] 2 Sep 2014

arxiv: v1 [cs.sy] 2 Sep 2014 Noname manuscrp No. wll be nsered by he edor Sgnalng for Decenralzed Roung n a Queueng Nework Y Ouyang Demoshens Tenekezs Receved: dae / Acceped: dae arxv:409.0887v [cs.sy] Sep 04 Absrac A dscree-me decenralzed

More information

Tight results for Next Fit and Worst Fit with resource augmentation

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

More information

Density Matrix Description of NMR BCMB/CHEM 8190

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

More information

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

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

More information

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

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

More information

Department of Economics University of Toronto

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

More information

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

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

More information

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

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

More information

January Examinations 2012

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

More information

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

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

More information

Planar truss bridge optimization by dynamic programming and linear programming

Planar truss bridge optimization by dynamic programming and linear programming IABSE-JSCE Jon Conference on Advances n Brdge Engneerng-III, Augus 1-, 015, Dhaka, Bangladesh. ISBN: 978-984-33-9313-5 Amn, Oku, Bhuyan, Ueda (eds.) www.abse-bd.org Planar russ brdge opmzaon by dynamc

More information

Advanced Machine Learning & Perception

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

More information

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

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

More information

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

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

More information

DEEP UNFOLDING FOR MULTICHANNEL SOURCE SEPARATION SUPPLEMENTARY MATERIAL

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

More information

THE PREDICTION OF COMPETITIVE ENVIRONMENT IN BUSINESS

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

More information

How about the more general "linear" scalar functions of scalars (i.e., a 1st degree polynomial of the following form with a constant term )?

How about the more general linear scalar functions of scalars (i.e., a 1st degree polynomial of the following form with a constant term )? lmcd Lnear ransformaon of a vecor he deas presened here are que general hey go beyond he radonal mar-vecor ype seen n lnear algebra Furhermore, hey do no deal wh bass and are equally vald for any se of

More information

M. Y. Adamu Mathematical Sciences Programme, AbubakarTafawaBalewa University, Bauchi, Nigeria

M. Y. Adamu Mathematical Sciences Programme, AbubakarTafawaBalewa University, Bauchi, Nigeria IOSR Journal of Mahemacs (IOSR-JM e-issn: 78-578, p-issn: 9-765X. Volume 0, Issue 4 Ver. IV (Jul-Aug. 04, PP 40-44 Mulple SolonSoluons for a (+-dmensonalhroa-sasuma shallow waer wave equaon UsngPanlevé-Bӓclund

More information

Density Matrix Description of NMR BCMB/CHEM 8190

Density Matrix Description of NMR BCMB/CHEM 8190 Densy Marx Descrpon of NMR BCMBCHEM 89 Operaors n Marx Noaon If we say wh one bass se, properes vary only because of changes n he coeffcens weghng each bass se funcon x = h< Ix > - hs s how we calculae

More information

MANY real-world applications (e.g. production

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

More information

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

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

More information

Genetic Algorithm in Parameter Estimation of Nonlinear Dynamic Systems

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

More information

Machine Learning Linear Regression

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

More information

Part II CONTINUOUS TIME STOCHASTIC PROCESSES

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

More information

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

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

More information

Existence and Uniqueness Results for Random Impulsive Integro-Differential Equation

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

More information

Solving the multi-period fixed cost transportation problem using LINGO solver

Solving the multi-period fixed cost transportation problem using LINGO solver Inernaonal Journal of Pure and Appled Mahemacs Volume 119 No. 12 2018, 2151-2157 ISSN: 1314-3395 (on-lne verson) url: hp://www.pam.eu Specal Issue pam.eu Solvng he mul-perod fxed cos ransporaon problem

More information

Single-loop System Reliability-Based Design & Topology Optimization (SRBDO/SRBTO): A Matrix-based System Reliability (MSR) Method

Single-loop System Reliability-Based Design & Topology Optimization (SRBDO/SRBTO): A Matrix-based System Reliability (MSR) Method 10 h US Naonal Congress on Compuaonal Mechancs Columbus, Oho 16-19, 2009 Sngle-loop Sysem Relably-Based Desgn & Topology Opmzaon (SRBDO/SRBTO): A Marx-based Sysem Relably (MSR) Mehod Tam Nguyen, Junho

More information

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

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

More information

Introduction to Boosting

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

More information

The preemptive resource-constrained project scheduling problem subject to due dates and preemption penalties: An integer programming approach

The preemptive resource-constrained project scheduling problem subject to due dates and preemption penalties: An integer programming approach Journal of Indusral Engneerng 1 (008) 35-39 The preempve resource-consraned projec schedulng problem subjec o due daes and preempon penales An neger programmng approach B. Afshar Nadjaf Deparmen of Indusral

More information

WiH Wei He

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

More information

UNIVERSITAT AUTÒNOMA DE BARCELONA MARCH 2017 EXAMINATION

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

More information

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

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

More information

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

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

More information

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

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

More information

Chapter Lagrangian Interpolation

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

More information

General Weighted Majority, Online Learning as Online Optimization

General Weighted Majority, Online Learning as Online Optimization Sascal Technques n Robocs (16-831, F10) Lecure#10 (Thursday Sepember 23) General Weghed Majory, Onlne Learnng as Onlne Opmzaon Lecurer: Drew Bagnell Scrbe: Nahanel Barshay 1 1 Generalzed Weghed majory

More information

A HIERARCHICAL KALMAN FILTER

A HIERARCHICAL KALMAN FILTER A HIERARCHICAL KALMAN FILER Greg aylor aylor Fry Consulng Acuares Level 8, 3 Clarence Sree Sydney NSW Ausrala Professoral Assocae, Cenre for Acuaral Sudes Faculy of Economcs and Commerce Unversy of Melbourne

More information

Epistemic Game Theory: Online Appendix

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

More information

2.1 Constitutive Theory

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

More information

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

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

More information

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

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

More information

Mohammad H. Al-Towaiq a & Hasan K. Al-Bzoor a a Department of Mathematics and Statistics, Jordan University of

Mohammad H. Al-Towaiq a & Hasan K. Al-Bzoor a a Department of Mathematics and Statistics, Jordan University of Ths arcle was downloaded by: [Jordan Unv. of Scence & Tech] On: 05 Aprl 05, A: 0:4 Publsher: Taylor & Francs Informa Ld Regsered n England and ales Regsered umber: 07954 Regsered offce: Mormer House, 37-4

More information

Relative controllability of nonlinear systems with delays in control

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

More information

Supplementary Material to: IMU Preintegration on Manifold for E cient Visual-Inertial Maximum-a-Posteriori Estimation

Supplementary Material to: IMU Preintegration on Manifold for E cient Visual-Inertial Maximum-a-Posteriori Estimation Supplemenary Maeral o: IMU Prenegraon on Manfold for E cen Vsual-Ineral Maxmum-a-Poseror Esmaon echncal Repor G-IRIM-CP&R-05-00 Chrsan Forser, Luca Carlone, Fran Dellaer, and Davde Scaramuzza May 0, 05

More information

FI 3103 Quantum Physics

FI 3103 Quantum Physics /9/4 FI 33 Quanum Physcs Aleander A. Iskandar Physcs of Magnesm and Phooncs Research Grou Insu Teknolog Bandung Basc Conces n Quanum Physcs Probably and Eecaon Value Hesenberg Uncerany Prncle Wave Funcon

More information

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

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

More information