Improving Anytime Point-Based Value Iteration Using Principled Point Selections

Size: px
Start display at page:

Download "Improving Anytime Point-Based Value Iteration Using Principled Point Selections"

Transcription

1 In In Proceedngs of the Twenteth Interntonl Jont Conference on Artfcl Intellgence (IJCAI-7) Improvng Anytme Pont-Bsed Vlue Iterton Usng Prncpled Pont Selectons Mchel R. Jmes, Mchel E. Smples, nd Dmtr A. Dolgov AI nd Robotcs Group Techncl Reserch, Toyot Techncl Center USA {mchel.jmes, mchel.smples, Abstrct Plnnng n prtlly-observble dynmcl systems (such s POMDPs nd PSRs) s computtonlly chllengng tsk. Populr pproxmton technques tht hve proved successful re pont-bsed plnnng methods ncludng pontbsed vlue terton (PBVI), whch works by pproxmtng the soluton t fnte set of ponts. These pont-bsed methods typclly re nytme lgorthms, whereby n ntl soluton s obtned usng smll set of ponts, nd the soluton my be ncrementlly mproved by ncludng ddtonl ponts. We ntroduce fmly of nytme PBVI lgorthms tht use the nformton present n the current soluton for dentfyng nd ddng new ponts tht hve the potentl to best mprove the next soluton. We motvte nd present two dfferent methods for choosng ponts nd evlute ther performnce emprclly, demonstrtng tht hgh-qulty solutons cn be obtned wth sgnfcntly fewer ponts thn prevous PBVI pproches. Introducton Pont-bsed plnnng lgorthms Pneu et l., 23; Spn nd Vlsss, 2] for prtlly-observble dynmcl systems hve become populr due to ther reltvely good performnce nd becuse they cn be mplemented s nytme lgorthms. It hs been suggested (e.g., n Pneu et l., 2]), tht nytme pont-bsed plnnng lgorthms could beneft sgnfcntly from the prncpled selecton of ponts to dd. We provde severl methods for usng nformton collected durng vlue terton (specfclly, chrcterstcs of the vlue functon) to choose new ddtonl ponts. We show tht propertes of the vlue functon llow the computton of n upper bound on the potentl mprovement (gn) resultng from the ddton of sngle pont. We rgue tht the ddton of ponts wth mxml gn produces good pproxmtons of the vlue functon, nd ths rgument s emprclly verfed. Further, we defne n optmzton problem tht fnds the pont wth the mxml gn for gven regon. These new pproches re emprclly compred to trdtonl nytme pont-bsed plnnng. The results show tht our lgorthms choose ddtonl ponts tht sgnfcntly mprove the resultng pproxmton of the vlue functon. Such mprovement s potentlly benefcl n two wys. Frst, computton tme cn be reduced becuse there re mny fewer ponts for whch computton must be performed. However, ths beneft s sometmes (but not lwys) offset by the ddtonl tme requred to select new ponts. Another potentl beneft s reduced memory storge for smller sets of ponts. Currently, the soluton of most systems do not requre lrge numbers of ponts, but storge beneft my become more mportnt s the sze of problems ncreses such tht more ponts re requred. 2 Bckground The plnnng methods developed here re for prtllyobservble, dscrete-tme, fnte-spce dynmcl systems n whch ctons re chosen from set A, observtons from set O, nd rewrds from set R. There s exctly one cton, one observton, nd one rewrd per tme-step plnnng lgorthms ttempt to fnd polcy tht mxmzes the longterm dscounted rewrd wth dscount fctor γ, ). These pont-bsed plnnng methods re sutble for t lest two clsses of models tht cn represent such dynmcl systems, the well-known prtlly observble Mrkov decson processes (POMDP), nd the newer predctve stte representton (PSR) Sngh et l., 24]. For the ske of fmlrty, we use POMDPs here, but t should be noted tht the methods presented here (s well s other plnnng methods for such dynmcl systems) cn just s esly be ppled to PSRs (see Jmes et l., 24] for detls). 2. POMDPs POMDPs re models of dynmcl systems bsed on underlyng ltent vrbles clled nomnl sttes s S. At ny tme step, the gent s n one nomnl stte s. Upon tkng n cton A the gent receves rewrd r(s, ) tht s functon of the stte nd cton, trnstons to new stte s ccordng to stochstc trnston model pr(s s, ), nd receves n observton o O ccordng to stochstc observton model pr(o s, ). For compctness, we defne the vector r s r (s) = r(s, ). The notton v(e) for vector v refers the vlue of entry e n v.

2 The gent does not drectly observe the nomnl stte s, nd so, must mntn some sort of memory to fully chrcterze ts current stte. In POMDPs, ths s typclly done usng the belef stte: probblty dstrbuton over nomnl sttes. The belef stte b summrzes the entre hstory by computng the probblty of beng n ech nomnl stte b(s). Ths computton of the new belef stte b o reched fter tkng cton nd gettng observton o from belef stte b s gven n the followng updte: b o = s pr(o s, ) s pr(s s, )b(s). pr(o, b) Ths equton s used s the gent ntercts wth the dynmcl system to mntn the gent s current belef stte. 2.2 Pont-bsed Vlue Iterton Algorthms Pont-bsed plnnng lgorthms, such s PBVI nd Perseus, for prtlly observble dynmcl systems perform plnnng by constructng n pproxmton of the vlue functon. By updtng the vlue functon for one pont, the vlue-functon pproxmton for nerby ponts s lso mproved. The bsc de s to clculte the vlue functon (nd ts grdent) only t smll number of ponts, nd the vlue functon for the other ponts wll be crred long. Ths pont-bsed pproch uses only frcton of the computtonl resources to construct the pproxmte vlue functon, s compred to exct methods. There re two mn sets of vectors used: the set P of ponts, ech of whch s belef stte vector; nd the set S n whch represents the pproxmte vlue functon V n t step n. The vlue functon s pecewse lner convex functon over belef sttes, defned s the upper surfce of the vectors α S n. The vlue for belef stte b s V n (b) = mx α S n b T α. The Bellmn equton s used to defne the updte for the next vlue functon: V n+ (b) = mx b T r + γ ] pr(o, b)v n (b o ) o = mx = mx γ o = mx where b T r + γ o b T r + pr(o, b) mx α S n (b o ) T α ] mx pr(o s, ) α S n s s b T r + γ ] mx {go o } bt go ] pr(s s, )b(s)α (s ) g o(s) = s pr(o s, )pr(s s, )α (s ), () for α S n. These g vectors re defned s such becuse they led to computtonl svngs. The computton of α vectors for V n+ (b) uses the so-clled bckup opertor. bckup(b) = rgmx b T g, b (2) {g b } A where g b = r + γ o rgmx {g o } bt g o. In PBVI, the bckup s ppled to ech pont n P, s follows. functon onesteppbvibckup(p, S old ). S new = 2. For ech b P compute α = bckup(b) nd set S new = S new α 3. return S new For Perseus, n ech terton the current set of α vectors s trnsformed to new set of α vectors by rndomly choosng ponts from the current set P nd pplyng the bckup opertor to them. Ths process s: functon onestepperseusbckup(p, S old ). S new =, P = P, where P lsts the non-mproved ponts 2. Smple b unformly t rndom from P, compute α = bckup(b) 3. If b T α V n (b) then dd α to S new, else dd α = rgmx α S old b T α to S new 4. Compute P = {b P : V n+ (b) ɛ < V n (b)}. If P s empty return S new, otherwse go to step 2 Gven ths, the bsc (non-nytme) lgorthms re, for * PBVI, Perseus]: functon bsc*(). P = pckintlponts(), S = mnmlalph(), n = 2. S n+ = onestep*bckup(p, S n ) 3. ncrement n, goto 2 where pckintlponts() typclly uses rndom wlk wth dstnce-bsed metrcs to choose n ntl set of ponts, nd the functon mnmlalph() returns the α vector wth ll entres equl to mn r R r/( γ). The lgorthm wll stop on ether condton on the vlue functon or on the number of tertons. Ths lgorthm s esly expnded to be n nytme lgorthm by modfyng pckintlponts() to strt wth smll set of ntl ponts nd ncludng step tht tertvely expnds the set of current ponts P. Typclly, hvng current set wth more ponts wll result n better pproxmton of the vlue functon, but wll requre more computton to updte. The nytme versons of the lgorthms re: functon nytme*(). P = pckintlponts(), S = mnmlalph(), n = 2. f(redytoaddponts()) then P = P bestpontstoadd(fndcnddteponts()) 3. S n+ = onestep*bckup(p, S n ) 4. ncrement n, goto 2

3 There re three new functons ntroduced n ths nytme lgorthm: redytoaddponts() whch determnes when the lgorthm dds new ponts to S; fndcnddteponts() whch s the frst step n determnng the ponts to dd; nd best- PontsToAdd() whch tkes the set of cnddte ponts nd determnes set of ponts to dd. In Secton 4, we present vrtons of these functons, ncludng our mn contrbutons, methods for determnng the best ponts to dd bsed on exmnton of the vlue functon. 3 Incrementl PBVI nd Perseus In ths secton we present methods for fndng the best ponts to dd, gven set of cnddte ponts. In some cses, the best ponts wll be subset of the cnddtes, nd n other cses new ponts tht re outsde the set of orgnl cnddtes wll be dentfed s best. In ll cses, we wll mke use of sclr mesure clled the gn of pont, whch s wy to evlute how useful tht pont wll be. We show how the current (pproxmte) vlue functon cn be used to compute useful mesures of gn tht cn then be used n nformed methods of pont selecton. For pont b n the belef spce, the gn g(b) estmtes how much the vlue functon wll mprove. The problem, then, s to construct mesure of gn tht dentfes the ponts tht wll most mprove the pproxmton of the vlue functon. Ths s complex problem, prmrly becuse chngng the vlue functon t one pont cn ffect the vlue of mny (or ll) other ponts, s cn be seen by exmnng the bckup opertor (Equton 2). Chnges my lso propgte over n rbtrry number of tme steps, s the chnges to other ponts propgte to yet other ponts. Therefore, exctly dentfyng the ponts wth best gn s much too computtonlly expensve, nd other mesures of gn must be used. Here, we present two dfferent mesures of gn: one bsed on exmnng the bckup opertor, wrtten g B, nd the other s derved from fndng n upper bound on the mount of gn tht pont my hve. We use lner progrmmng to compute ths gn, wrtten g LP. Gven these defntons of gn, we present two dfferent methods for dentfyng ponts to dd. The smplest method (detled n Secton 4) tkes the cnddte ponts, orders them ccordng to ther gns, nd then selects the top few. A more sophstcted method leverges some nformton present when computng g LP to dentfy ponts tht hve even hgher gn thn the cnddte ponts, nd so returns dfferent ponts thn the orgnl cnddtes. We descrbe how these selecton methods cn be mplemented n Secton 4, followng detled dscusson (Sectons 3. nd 3.2) of the two mesures of gn. 3. Gn Bsed on One-Step Bckup The gn g B mesures the dfference between the current vlue of pont b nd ts vlue ccordng to n α vector computed by one-step bckup of b v Equton 2. Ths mesure does not hve strong theoretcl support, but ntutvely, f the one-step dfference for pont s lrge, then t wll lso mprove the pproxmtons of nerby ponts sgnfcntly. These mprovements wll then hve postve effects on ponts tht led to these ponts (lthough the effect wll be dscounted), mprovng mny prts of the vlue functon. Ths gn s defned s: g B (b) = mx r(b, ) + γ o = b T α V (b), ] pr(o b, )V (b o ) V (b) where α = bckup (b). Note tht ths mesure ncludes the mmedte expected rewrd t b, quntty tht ws not prevously ncluded n the computton of ny α vector. Ths mesure cn mke use of cched g vectors (Equton ), nd so t s nexpensve from the computtonl stndpont. In the followng secton, we descrbe dfferent mesure of gn tht hs stronger theoretcl justfcton, but lso comes t the expensve of hgher computtonl cost. 3.2 Gn Bsed on Vlue-Functon Bounds Ths second mesure of gn tht we ntroduce uses the followng nsght, whch we frst llustrte on system wth two nomnl sttes, then extend the ntuton to three-stte system, nd then descrbe the generl clculton for ny number of sttes. The vlue functon s pecewse lner convex functon consstng of (for two nomnl sttes) the upper surfce of set of lnes defned by α vectors (see Fgure for runnng exmple). The vectors whch defne these lnes correspond to ponts nteror to the regons where the correspondng lnes re mxml (ponts A, B, nd C). Intutvely, the vlue functon s most relevnt t these ponts, nd becomes less relevnt wth ncresng dstnce from the pont. Here, we ssume tht the vlue gven by the α vector s correct t the correspondng pont. Now consder the ncluson of new α vector for new pont (pont X n Fgure ). If the vlue functon for pont A s correct, then the new α vector cnnot chnge the vlue of A to be greter thn ts current vlue; the sme holds for B. Therefore, the new α vector for pont X s upper-bounded by the lne between the vlue for A nd the vlue for B. Thus, the gn for X s the dfference between ths upper bound nd the current vlue of X. For the cse wth three nomnl sttes, the nlogous plnr soluton to the upper bound s not s smple to compute (see Fgure b nd c). To vsulze the problem, consder moble plne beng pushed upwrd from pont X, whch s constrned only to not move bove ny of the exstng ponts tht defne the vlue-functon surfce (A, B, C, nd D n Fgure b). In the exmple depcted n Fgure b, ths plne wll come to rest on the ponts defned by the loctons of A, B, C n the belef spce nd ther vlues. The vlue of tht plne t X then gves the tghtest upper bound for ts vlue f we were to dd new α vector t X. The gn s then computed n the sme mnner s n the two stte cse bove, but fndng the bove-descrbed plne s not strghtforwrd (nd becomes even more dffcult n hgherdmensonl spces). The dffculty s tht n hgher dmensonl spces n contrst to the two stte cse where the mnml upper bound s lwys defned by the ponts mmedtely to the left nd rght of the new pont selectng the ponts tht defne tht mnml upper-boundng surfce s non-trvl tsk. However, ths problem cn be formulted s

4 Vlue g LP (X) A X B C p(s ) () (b) (c) Fgure : (): Vlue functon for dynmcl system wth two nomnl sttes (note tht p(s 2 ) = p(s ), so n xs for p(s 2 ) s unnecessry). The vlue functon s defned by the α vectors correspondng to ponts A, B, nd C. The gn g LP (X) for pont X s the dfference between the upper bound defned by the lne V (A), V (B) nd the current vlue of X. (b,c): Vlue functon for system wth three nomnl sttes. The vlue functon s defned by the α vectors correspondng to ponts A, B, C, nd D, nd the gn g LP (X) for pont X s the dfference between the tghtest upper bound (defned by plne V (A), V (B), V (C) ) nd the current vlue for X. Tht tghtest upper-bound plne s produced by the frst LP (FndGnAndRegonForPont). X s the best pont produced by the second LP (FndBestPontForRegon) for the boundng regon {A, B, C}. lner progrm (LP). We descrbe ths LP n Tble nd refer to t s FndGnAndRegonForPont(b ). Gven cnddte belef pont b ts current sclr vlue v, the current ponts {b...b m }, the current sclr vlues v...v m of these ponts, nd the vlue vectors α...α n, ths mnmzton LP fnds the tghtest upper bound V u (b ) = w v, whch cn then be used to compute the gn g LP (b ) = V u (b ) V (b ). In words, ths mnmzton LP works by fndng set of ponts (these re the set of ponts wth nonzero weghts w ) such tht: ) b cn be expressed convex lner combnton of these ponts, nd ) the vlue functon defned by the plne tht rests on ther vlue-functon ponts s the mnml surfce bove v. We cll the set of such ponts wth non-zero weghts the boundng regon Γ(b ) = {b w > } for pont b ; t hs the property tht the pont b must be nteror to t, nd these boundng regons wll ply n mportnt role n the next secton. There s one subtle ssue regrdng the bove LP, whch s tht not ll ponts of potentl nterest re nteror to the set of current ponts for whch α vectors re defned (n fct erly n the process when there re few ponts, most of the useful ponts wll be outsde the convex hull of current ponts). Ths s problemtc becuse we wnt to consder these ponts (they often correspond to reltvely poorly pproxmted res of the vlue functon), but they re never nteror to ny subset of the current ponts. To resolve ths ssue, we ntroduce the set U of unt bss ponts: the ponts wth ll components but one equl to, nd the remnng entry equl to (the stndrd orthonorml bss n the belef spce). We then ugment the set of current belef ponts n our LP to nclude the unt bss ponts: {b...b m } = {P U}. The vlues v j of the ponts n U \ P re computed s v j = bckup (b j ) b j U \ P, snce these ponts were not updted n onestep*bckup() s were the ponts n P. As before, the vrbles n ths ugmented LP re the sclr weghts w...w m tht defne whch ponts from {P U} form the boundng regon Γ(b ) 3.3 Fndng the Best Pont n Boundng Regon Gven the gns g LP for the cnddte ponts, t s possble to choose ponts to dd bsed on ths crter, but one drwbck to ths s tht we mght dd multple ponts from the sme boundng regon Γ, nd thus wste resources by pproxmtng the vlue functon for multple ponts when just one would work lmost s well. Further, t my be tht none of these ponts s the pont n the regon wth the hghest gn. To ddress these ssues, we ntroduce nother LP clled FndBestPontForRegon(Γ) (Tble ) tht tkes boundng regon nd fnds the pont nteror to tht regon wth mxml g LP. Note tht ths regon my contn unt bss ponts, llowng the new pont to be exteror to the current set of ponts. Ths LP tkes s nput boundng regon Γ(b ) = {b...b l }, sclr vlues v...v l of these ponts, nd the vlue vectors α...α n. The optmzton vrbles re: sclr weghts w...w l, vector b new, whch s the pont to be found, nd the sclr vlue v new, whch s the vlue of b new. The process of fndng the best ponts to dd wll cll ths LP on ll unque regons found by runnng the frst LP (FndGnAndRegonForPont) on ll cnddte ponts. 4 Usng Gns n PBVI nd Perseus We now defne two ncrementl lgorthms, PBVI-I nd Perseus-I, usng the gns g B nd g LP, nd the lner progrms bove. Referencng the nytme*() lgorthm, we explored vrous possbltes for redytoaddponts(), bestpontstoadd(c), nd fndcnddteponts(). Whle mny vrtons exst, ths pper presents only few tht seem most promsng.

5 Tble : The lner progrms used to ) compute the gn nd regon Γ(b ) for pont b gven vlue functon, nd ) fnd the best pont wthn gven regon Γ gven the vlue functon. FndGnAndRegonForPont(b ): Mnmze: w v Gven: Vrbles: Constrnts: b, v, b...b m, v...v m, α...α n w...w m b = w b ; w v v w = ; w..m] Two computtonlly chep possbltes for redytoadd- Ponts() re to dd ponts: ) fter fxed number of tertons, nd ) when the updte process hs pproxmtely converged,.e., the mxml one-step dfference n vlue over ll ponts s below some threshold (mx b P (V n+ (b) V n (b)) ɛ). We hve lso explored two methods for fndcnddteponts(). The frst s to keep lst of ponts tht re one-step successors to the ponts n P, the set {b o b P }. The second method s to do rndom wlk nd use dstnce threshold to choose ponts, just s n pckintlponts(). The gns dscussed n the prevous secton cn then be used for bestpontstoadd(c). Gn g B orders the cnddte ponts C nd the top k re chosen. For gn g LP, the cnddte ponts defne set of unque boundng regons {Γ(b) b C}. For ech regon, the LP FndBestPontFor- Regon() returns the best cnddte pont b. These cnddte ponts re ordered by g LP (b ) nd the top k re chosen. To compre our methods gnst bselne nytme lgorthm, we used the stndrd rndom wlk method, choosng the frst k ponts tht exceeded dstnce threshold. Experments To evlute our prncpled pont selecton lgorthms, we conducted testng on set of stndrd POMDP domns. Three of the problems re smll-to-md szed, nd the fourth (Hllwy) s lrger. Domn defntons my be obtned t Cssndr, 999]. We tested three lgorthms: ) bselne, nytme Perseus usng the dstnce-bsed metrc; ) one-step bckup Perseus-I usng g B ; nd ) LP-bsed Perseus-I usng g LP nd the two LPs for fndng the best ponts. All three lgorthms used the convergence condton wth ɛ = E 3 for redytoaddponts(). The lgorthms for Network, Shuttle, nd 4x3 dded up to 3 ponts t tme, whle Hllwy dded up to 3 ponts t tme. However, choosng ponts usng LPs often dd not dd ll 3 becuse fewer unque regons were found. In the one-step bckup Perseus- I, we used the successor metrc for fndcnddteponts() s t ws esly ntegrted wth lttle overhed, whle n the LPbsed Perseus-I we used rndom wlk wth dstnce metrc s t ntroduced the lest overhed n ths cse. The results were verged over 3 runs, nd the performnce metrc s the verge dscounted rewrd, for gven ntl belef stte. The results re presented n the grphs n Fgure 2, whch re of FndBestPontForRegon(Γ) ( ) Mxmze: w v v new Gven: Vrbles: Constrnts: b...b l, v...v l, α...α n w...w l, v new, b new b new = w b ; w..l] w = ; v new α j b new j..l] three types. The frst type (performnce vs. number of ponts) s mesure of the effectveness of pont selecton. The second type (performnce vs. number tertons) nd the thrd type (performnce vs. CPU tme) both mesure how well the use of prncpled pont selecton ffects the overll performnce, but the thrd type shows the mpct of ddtonl tme ncurred by the pont-selecton process. These grphs show performnce on the x-xs, nd the mount of resource (ponts, tertons, or tme) used to cheve tht performnce on the y-xs. When comprng two lgorthms wth respect to gven resource, the better lgorthm wll be lower. Exmnton of the frst-type grphs shows tht the LPbsed Perseus-I dd good job of dentfyng the best ponts to dd. The resultng polces produced better performnce wth fewer ponts thn ether of the other two methods. Furthermore, on three of the problems, the one-step bckup Perseus- I lso dd sgnfcntly better thn the bselne for choosng ponts. Thus, for ths metrc, our gol of fndng useful ponts hs been cheved. Most promsngly, on the lrgest problem (Hllwy) our lgorthms used sgnfcntly fewer ponts to cheve good performnce. However, the other grphs show tht choosng good ponts does not lwys trnslte nto fster lgorthm. On one problem (Network) both versons of Perseus-I were fster thn the bselne, but on the other problems, the performnce ws ether closely compettve or slghtly worse due to the overhed ncurred by dentfyng good ponts. Whle ths result ws not optml, the fct tht there were sgnfcnt speedups on one problem combned wth the pont selecton results leds us to drw the concluson tht ths method s promsng nd needs further nvestgton nd development. References Cssndr, 999] A. Cssndr. Tony s pomdp pge. pomdp/ndex.html, 999. Izd et l., 2] Msoumeh T. Izd, Ajt V. Rjwde, nd Don Precup. Usng core belefs for pont-bsed vlue terton. In 9th Interntonl Jont Conference on Artfcl Intellgence, 2. Jmes et l., 24] Mchel R. Jmes, Stnder Sngh, nd Mchel L. Lttmn. Plnnng wth predctve stte representtons. In The 24 Interntonl Conference on Mchne Lernng nd Applctons, 24.

6 Ponts vs. Performnce Itertons vs. Performnce 6 Tme vs. Performnce 2 LP 4 Dstnce 4 One Step 3. Network Shuttle x Hllwy Fgure 2: Emprcl comprsons of three methods for selectng new ponts durng vlue terton. Ech grph evlutes performnce gnst one of three metrcs number of ponts, totl number of tertons used n constructng polcy, nd totl CPU tme. The y-xs of ech grph represents the verge vlue of the column s metrc requred to ern prtculr rewrd (x-xs) durng emprcl evluton. Experments were conducted on four dynmcl systems (one per row): Network, Shuttle, 4x3, nd Hllwy. Lower vlues re ndctve of less resource consumpton, nd so re more desrble. Pneu et l., 23] J. Pneu, G. Gordon, nd S. Thrun. Pontbsed vlue terton: An nytme lgorthm for POMDPs. In Proc. 8th Interntonl Jont Conference on Artfcl Intellgence, Acpulco, Mexco, 23. Pneu et l., 2] J. Pneu, G. Gordon, nd S. Thrun. Pontbsed pproxmtons for fst pomdp solvng. Techncl Report SOCS-TR-2.4, McGll Unversty, 2. Shn et l., 2] Guy Shn, Ronen Brfmn, nd Solomon Shmony. Model-bsed onlne lernng of POMDPs. In 6th Euro- pen Conference on Mchne Lernng, 2. Sngh et l., 24] Stnder Sngh, Mchel R. Jmes, nd Mtthew R. Rudry. Predctve stte representtons, new theory for modelng dynmcl systems. In 2th Conference on Uncertnty n Artfcl Intellgence, 24. Spn nd Vlsss, 2] M.T.J Spn nd N. Vlsss. Perseus: Rndomzed pont-bsed vlue terton for pomdps. Journl of Artfcl Intellgence Reserch, 24:9 22, 2.

Partially Observable Systems. 1 Partially Observable Markov Decision Process (POMDP) Formalism

Partially Observable Systems. 1 Partially Observable Markov Decision Process (POMDP) Formalism CS294-40 Lernng for Rootcs nd Control Lecture 10-9/30/2008 Lecturer: Peter Aeel Prtlly Oservle Systems Scre: Dvd Nchum Lecture outlne POMDP formlsm Pont-sed vlue terton Glol methods: polytree, enumerton,

More information

Chapter Newton-Raphson Method of Solving a Nonlinear Equation

Chapter Newton-Raphson Method of Solving a Nonlinear Equation Chpter.4 Newton-Rphson Method of Solvng Nonlner Equton After redng ths chpter, you should be ble to:. derve the Newton-Rphson method formul,. develop the lgorthm of the Newton-Rphson method,. use the Newton-Rphson

More information

Dennis Bricker, 2001 Dept of Industrial Engineering The University of Iowa. MDP: Taxi page 1

Dennis Bricker, 2001 Dept of Industrial Engineering The University of Iowa. MDP: Taxi page 1 Denns Brcker, 2001 Dept of Industrl Engneerng The Unversty of Iow MDP: Tx pge 1 A tx serves three djcent towns: A, B, nd C. Ech tme the tx dschrges pssenger, the drver must choose from three possble ctons:

More information

Rank One Update And the Google Matrix by Al Bernstein Signal Science, LLC

Rank One Update And the Google Matrix by Al Bernstein Signal Science, LLC Introducton Rnk One Updte And the Google Mtrx y Al Bernsten Sgnl Scence, LLC www.sgnlscence.net here re two dfferent wys to perform mtrx multplctons. he frst uses dot product formulton nd the second uses

More information

Applied Statistics Qualifier Examination

Applied Statistics Qualifier Examination Appled Sttstcs Qulfer Exmnton Qul_june_8 Fll 8 Instructons: () The exmnton contns 4 Questons. You re to nswer 3 out of 4 of them. () You my use ny books nd clss notes tht you mght fnd helpful n solvng

More information

UNIVERSITY OF IOANNINA DEPARTMENT OF ECONOMICS. M.Sc. in Economics MICROECONOMIC THEORY I. Problem Set II

UNIVERSITY OF IOANNINA DEPARTMENT OF ECONOMICS. M.Sc. in Economics MICROECONOMIC THEORY I. Problem Set II Mcroeconomc Theory I UNIVERSITY OF IOANNINA DEPARTMENT OF ECONOMICS MSc n Economcs MICROECONOMIC THEORY I Techng: A Lptns (Note: The number of ndctes exercse s dffculty level) ()True or flse? If V( y )

More information

Remember: Project Proposals are due April 11.

Remember: Project Proposals are due April 11. Bonformtcs ecture Notes Announcements Remember: Project Proposls re due Aprl. Clss 22 Aprl 4, 2002 A. Hdden Mrov Models. Defntons Emple - Consder the emple we tled bout n clss lst tme wth the cons. However,

More information

4. Eccentric axial loading, cross-section core

4. Eccentric axial loading, cross-section core . Eccentrc xl lodng, cross-secton core Introducton We re strtng to consder more generl cse when the xl force nd bxl bendng ct smultneousl n the cross-secton of the br. B vrtue of Snt-Vennt s prncple we

More information

Chapter Newton-Raphson Method of Solving a Nonlinear Equation

Chapter Newton-Raphson Method of Solving a Nonlinear Equation Chpter 0.04 Newton-Rphson Method o Solvng Nonlner Equton Ater redng ths chpter, you should be ble to:. derve the Newton-Rphson method ormul,. develop the lgorthm o the Newton-Rphson method,. use the Newton-Rphson

More information

Principle Component Analysis

Principle Component Analysis Prncple Component Anlyss Jng Go SUNY Bufflo Why Dmensonlty Reducton? We hve too mny dmensons o reson bout or obtn nsghts from o vsulze oo much nose n the dt Need to reduce them to smller set of fctors

More information

Introduction to Numerical Integration Part II

Introduction to Numerical Integration Part II Introducton to umercl Integrton Prt II CS 75/Mth 75 Brn T. Smth, UM, CS Dept. Sprng, 998 4/9/998 qud_ Intro to Gussn Qudrture s eore, the generl tretment chnges the ntegrton prolem to ndng the ntegrl w

More information

18.7 Artificial Neural Networks

18.7 Artificial Neural Networks 310 18.7 Artfcl Neurl Networks Neuroscence hs hypotheszed tht mentl ctvty conssts prmrly of electrochemcl ctvty n networks of brn cells clled neurons Ths led McCulloch nd Ptts to devse ther mthemtcl model

More information

DCDM BUSINESS SCHOOL NUMERICAL METHODS (COS 233-8) Solutions to Assignment 3. x f(x)

DCDM BUSINESS SCHOOL NUMERICAL METHODS (COS 233-8) Solutions to Assignment 3. x f(x) DCDM BUSINESS SCHOOL NUMEICAL METHODS (COS -8) Solutons to Assgnment Queston Consder the followng dt: 5 f() 8 7 5 () Set up dfference tble through fourth dfferences. (b) Wht s the mnmum degree tht n nterpoltng

More information

An Introduction to Support Vector Machines

An Introduction to Support Vector Machines An Introducton to Support Vector Mchnes Wht s good Decson Boundry? Consder two-clss, lnerly seprble clssfcton problem Clss How to fnd the lne (or hyperplne n n-dmensons, n>)? Any de? Clss Per Lug Mrtell

More information

International Journal of Pure and Applied Sciences and Technology

International Journal of Pure and Applied Sciences and Technology Int. J. Pure Appl. Sc. Technol., () (), pp. 44-49 Interntonl Journl of Pure nd Appled Scences nd Technolog ISSN 9-67 Avlle onlne t www.jopst.n Reserch Pper Numercl Soluton for Non-Lner Fredholm Integrl

More information

Dynamic Power Management in a Mobile Multimedia System with Guaranteed Quality-of-Service

Dynamic Power Management in a Mobile Multimedia System with Guaranteed Quality-of-Service Dynmc Power Mngement n Moble Multmed System wth Gurnteed Qulty-of-Servce Qnru Qu, Qng Wu, nd Mssoud Pedrm Dept. of Electrcl Engneerng-Systems Unversty of Southern Clforn Los Angeles CA 90089 Outlne! Introducton

More information

Variable time amplitude amplification and quantum algorithms for linear algebra. Andris Ambainis University of Latvia

Variable time amplitude amplification and quantum algorithms for linear algebra. Andris Ambainis University of Latvia Vrble tme mpltude mplfcton nd quntum lgorthms for lner lgebr Andrs Ambns Unversty of Ltv Tlk outlne. ew verson of mpltude mplfcton;. Quntum lgorthm for testng f A s sngulr; 3. Quntum lgorthm for solvng

More information

INTRODUCTION TO COMPLEX NUMBERS

INTRODUCTION TO COMPLEX NUMBERS INTRODUCTION TO COMPLEX NUMBERS The numers -4, -3, -, -1, 0, 1,, 3, 4 represent the negtve nd postve rel numers termed ntegers. As one frst lerns n mddle school they cn e thought of s unt dstnce spced

More information

Quiz: Experimental Physics Lab-I

Quiz: Experimental Physics Lab-I Mxmum Mrks: 18 Totl tme llowed: 35 mn Quz: Expermentl Physcs Lb-I Nme: Roll no: Attempt ll questons. 1. In n experment, bll of mss 100 g s dropped from heght of 65 cm nto the snd contner, the mpct s clled

More information

6 Roots of Equations: Open Methods

6 Roots of Equations: Open Methods HK Km Slghtly modfed 3//9, /8/6 Frstly wrtten t Mrch 5 6 Roots of Equtons: Open Methods Smple Fed-Pont Iterton Newton-Rphson Secnt Methods MATLAB Functon: fzero Polynomls Cse Study: Ppe Frcton Brcketng

More information

GAUSS ELIMINATION. Consider the following system of algebraic linear equations

GAUSS ELIMINATION. Consider the following system of algebraic linear equations Numercl Anlyss for Engneers Germn Jordnn Unversty GAUSS ELIMINATION Consder the followng system of lgebrc lner equtons To solve the bove system usng clsscl methods, equton () s subtrcted from equton ()

More information

523 P a g e. is measured through p. should be slower for lesser values of p and faster for greater values of p. If we set p*

523 P a g e. is measured through p. should be slower for lesser values of p and faster for greater values of p. If we set p* R. Smpth Kumr, R. Kruthk, R. Rdhkrshnn / Interntonl Journl of Engneerng Reserch nd Applctons (IJERA) ISSN: 48-96 www.jer.com Vol., Issue 4, July-August 0, pp.5-58 Constructon Of Mxed Smplng Plns Indexed

More information

Definition of Tracking

Definition of Tracking Trckng Defnton of Trckng Trckng: Generte some conclusons bout the moton of the scene, objects, or the cmer, gven sequence of mges. Knowng ths moton, predct where thngs re gong to project n the net mge,

More information

The Schur-Cohn Algorithm

The Schur-Cohn Algorithm Modelng, Estmton nd Otml Flterng n Sgnl Processng Mohmed Njm Coyrght 8, ISTE Ltd. Aendx F The Schur-Cohn Algorthm In ths endx, our m s to resent the Schur-Cohn lgorthm [] whch s often used s crteron for

More information

Lecture 4: Piecewise Cubic Interpolation

Lecture 4: Piecewise Cubic Interpolation Lecture notes on Vrtonl nd Approxmte Methods n Appled Mthemtcs - A Perce UBC Lecture 4: Pecewse Cubc Interpolton Compled 6 August 7 In ths lecture we consder pecewse cubc nterpolton n whch cubc polynoml

More information

Review of linear algebra. Nuno Vasconcelos UCSD

Review of linear algebra. Nuno Vasconcelos UCSD Revew of lner lgebr Nuno Vsconcelos UCSD Vector spces Defnton: vector spce s set H where ddton nd sclr multplcton re defned nd stsf: ) +( + ) (+ )+ 5) λ H 2) + + H 6) 3) H, + 7) λ(λ ) (λλ ) 4) H, - + 8)

More information

Computing a complete histogram of an image in Log(n) steps and minimum expected memory requirements using hypercubes

Computing a complete histogram of an image in Log(n) steps and minimum expected memory requirements using hypercubes Computng complete hstogrm of n mge n Log(n) steps nd mnmum expected memory requrements usng hypercubes TAREK M. SOBH School of Engneerng, Unversty of Brdgeport, Connectcut, USA. Abstrct Ths work frst revews

More information

THE COMBINED SHEPARD ABEL GONCHAROV UNIVARIATE OPERATOR

THE COMBINED SHEPARD ABEL GONCHAROV UNIVARIATE OPERATOR REVUE D ANALYSE NUMÉRIQUE ET DE THÉORIE DE L APPROXIMATION Tome 32, N o 1, 2003, pp 11 20 THE COMBINED SHEPARD ABEL GONCHAROV UNIVARIATE OPERATOR TEODORA CĂTINAŞ Abstrct We extend the Sheprd opertor by

More information

Math 497C Sep 17, Curves and Surfaces Fall 2004, PSU

Math 497C Sep 17, Curves and Surfaces Fall 2004, PSU Mth 497C Sep 17, 004 1 Curves nd Surfces Fll 004, PSU Lecture Notes 3 1.8 The generl defnton of curvture; Fox-Mlnor s Theorem Let α: [, b] R n be curve nd P = {t 0,...,t n } be prtton of [, b], then the

More information

A Tri-Valued Belief Network Model for Information Retrieval

A Tri-Valued Belief Network Model for Information Retrieval December 200 A Tr-Vlued Belef Networ Model for Informton Retrevl Fernndo Ds-Neves Computer Scence Dept. Vrgn Polytechnc Insttute nd Stte Unversty Blcsburg, VA 24060. IR models t Combnng Evdence Grphcl

More information

Two Coefficients of the Dyson Product

Two Coefficients of the Dyson Product Two Coeffcents of the Dyson Product rxv:07.460v mth.co 7 Nov 007 Lun Lv, Guoce Xn, nd Yue Zhou 3,,3 Center for Combntorcs, LPMC TJKLC Nnk Unversty, Tnjn 30007, P.R. Chn lvlun@cfc.nnk.edu.cn gn@nnk.edu.cn

More information

On-line Reinforcement Learning Using Incremental Kernel-Based Stochastic Factorization

On-line Reinforcement Learning Using Incremental Kernel-Based Stochastic Factorization On-lne Renforcement Lernng Usng Incrementl Kernel-Bsed Stochstc Fctorzton André M. S. Brreto School of Computer Scence McGll Unversty Montrel, Cnd msb@cs.mcgll.c Don Precup School of Computer Scence McGll

More information

Jens Siebel (University of Applied Sciences Kaiserslautern) An Interactive Introduction to Complex Numbers

Jens Siebel (University of Applied Sciences Kaiserslautern) An Interactive Introduction to Complex Numbers Jens Sebel (Unversty of Appled Scences Kserslutern) An Interctve Introducton to Complex Numbers 1. Introducton We know tht some polynoml equtons do not hve ny solutons on R/. Exmple 1.1: Solve x + 1= for

More information

CS434a/541a: Pattern Recognition Prof. Olga Veksler. Lecture 9

CS434a/541a: Pattern Recognition Prof. Olga Veksler. Lecture 9 CS434/541: Pttern Recognton Prof. Olg Veksler Lecture 9 Announcements Fnl project proposl due Nov. 1 1-2 prgrph descrpton Lte Penlt: s 1 pont off for ech d lte Assgnment 3 due November 10 Dt for fnl project

More information

ESCI 342 Atmospheric Dynamics I Lesson 1 Vectors and Vector Calculus

ESCI 342 Atmospheric Dynamics I Lesson 1 Vectors and Vector Calculus ESI 34 tmospherc Dnmcs I Lesson 1 Vectors nd Vector lculus Reference: Schum s Outlne Seres: Mthemtcl Hndbook of Formuls nd Tbles Suggested Redng: Mrtn Secton 1 OORDINTE SYSTEMS n orthonorml coordnte sstem

More information

Electrochemical Thermodynamics. Interfaces and Energy Conversion

Electrochemical Thermodynamics. Interfaces and Energy Conversion CHE465/865, 2006-3, Lecture 6, 18 th Sep., 2006 Electrochemcl Thermodynmcs Interfces nd Energy Converson Where does the energy contrbuton F zϕ dn come from? Frst lw of thermodynmcs (conservton of energy):

More information

Pyramid Algorithms for Barycentric Rational Interpolation

Pyramid Algorithms for Barycentric Rational Interpolation Pyrmd Algorthms for Brycentrc Rtonl Interpolton K Hormnn Scott Schefer Astrct We present new perspectve on the Floter Hormnn nterpolnt. Ths nterpolnt s rtonl of degree (n, d), reproduces polynomls of degree

More information

In this Chapter. Chap. 3 Markov chains and hidden Markov models. Probabilistic Models. Example: CpG Islands

In this Chapter. Chap. 3 Markov chains and hidden Markov models. Probabilistic Models. Example: CpG Islands In ths Chpter Chp. 3 Mrov chns nd hdden Mrov models Bontellgence bortory School of Computer Sc. & Eng. Seoul Ntonl Unversty Seoul 5-74, Kore The probblstc model for sequence nlyss HMM (hdden Mrov model)

More information

Bi-level models for OD matrix estimation

Bi-level models for OD matrix estimation TNK084 Trffc Theory seres Vol.4, number. My 2008 B-level models for OD mtrx estmton Hn Zhng, Quyng Meng Abstrct- Ths pper ntroduces two types of O/D mtrx estmton model: ME2 nd Grdent. ME2 s mxmum-entropy

More information

CONTEXTUAL MULTI-ARMED BANDIT ALGORITHMS FOR PERSONALIZED LEARNING ACTION SELECTION. Indu Manickam, Andrew S. Lan, and Richard G.

CONTEXTUAL MULTI-ARMED BANDIT ALGORITHMS FOR PERSONALIZED LEARNING ACTION SELECTION. Indu Manickam, Andrew S. Lan, and Richard G. CONTEXTUAL MULTI-ARMED BANDIT ALGORITHMS FOR PERSONALIZED LEARNING ACTION SELECTION Indu Mnckm, Andrew S. Ln, nd Rchrd G. Brnuk Rce Unversty ABSTRACT Optmzng the selecton of lernng resources nd prctce

More information

Model Fitting and Robust Regression Methods

Model Fitting and Robust Regression Methods Dertment o Comuter Engneerng Unverst o Clorn t Snt Cruz Model Fttng nd Robust Regresson Methods CMPE 64: Imge Anlss nd Comuter Vson H o Fttng lnes nd ellses to mge dt Dertment o Comuter Engneerng Unverst

More information

arxiv: v2 [cs.lg] 9 Nov 2017

arxiv: v2 [cs.lg] 9 Nov 2017 Renforcement Lernng under Model Msmtch Aurko Roy 1, Hun Xu 2, nd Sebstn Pokutt 2 rxv:1706.04711v2 cs.lg 9 Nov 2017 1 Google Eml: urkor@google.com 2 ISyE, Georg Insttute of Technology, Atlnt, GA, USA. Eml:

More information

Physics 121 Sample Common Exam 2 Rev2 NOTE: ANSWERS ARE ON PAGE 7. Instructions:

Physics 121 Sample Common Exam 2 Rev2 NOTE: ANSWERS ARE ON PAGE 7. Instructions: Physcs 121 Smple Common Exm 2 Rev2 NOTE: ANSWERS ARE ON PAGE 7 Nme (Prnt): 4 Dgt ID: Secton: Instructons: Answer ll 27 multple choce questons. You my need to do some clculton. Answer ech queston on the

More information

Machine Learning Support Vector Machines SVM

Machine Learning Support Vector Machines SVM Mchne Lernng Support Vector Mchnes SVM Lesson 6 Dt Clssfcton problem rnng set:, D,,, : nput dt smple {,, K}: clss or lbel of nput rget: Construct functon f : X Y f, D Predcton of clss for n unknon nput

More information

Katholieke Universiteit Leuven Department of Computer Science

Katholieke Universiteit Leuven Department of Computer Science Updte Rules for Weghted Non-negtve FH*G Fctorzton Peter Peers Phlp Dutré Report CW 440, Aprl 006 Ktholeke Unverstet Leuven Deprtment of Computer Scence Celestjnenln 00A B-3001 Heverlee (Belgum) Updte Rules

More information

Lecture 36. Finite Element Methods

Lecture 36. Finite Element Methods CE 60: Numercl Methods Lecture 36 Fnte Element Methods Course Coordntor: Dr. Suresh A. Krth, Assocte Professor, Deprtment of Cvl Engneerng, IIT Guwht. In the lst clss, we dscussed on the ppromte methods

More information

Investigation phase in case of Bragg coupling

Investigation phase in case of Bragg coupling Journl of Th-Qr Unversty No.3 Vol.4 December/008 Investgton phse n cse of Brgg couplng Hder K. Mouhmd Deprtment of Physcs, College of Scence, Th-Qr, Unv. Mouhmd H. Abdullh Deprtment of Physcs, College

More information

CHAPTER - 7. Firefly Algorithm based Strategic Bidding to Maximize Profit of IPPs in Competitive Electricity Market

CHAPTER - 7. Firefly Algorithm based Strategic Bidding to Maximize Profit of IPPs in Competitive Electricity Market CHAPTER - 7 Frefly Algorthm sed Strtegc Bddng to Mxmze Proft of IPPs n Compettve Electrcty Mrket 7. Introducton The renovton of electrc power systems plys mjor role on economc nd relle operton of power

More information

Least squares. Václav Hlaváč. Czech Technical University in Prague

Least squares. Václav Hlaváč. Czech Technical University in Prague Lest squres Václv Hlváč Czech echncl Unversty n Prgue hlvc@fel.cvut.cz http://cmp.felk.cvut.cz/~hlvc Courtesy: Fred Pghn nd J.P. Lews, SIGGRAPH 2007 Course; Outlne 2 Lner regresson Geometry of lest-squres

More information

Statistics and Probability Letters

Statistics and Probability Letters Sttstcs nd Probblty Letters 79 (2009) 105 111 Contents lsts vlble t ScenceDrect Sttstcs nd Probblty Letters journl homepge: www.elsever.com/locte/stpro Lmtng behvour of movng verge processes under ϕ-mxng

More information

Statistics 423 Midterm Examination Winter 2009

Statistics 423 Midterm Examination Winter 2009 Sttstcs 43 Mdterm Exmnton Wnter 009 Nme: e-ml: 1. Plese prnt your nme nd e-ml ddress n the bove spces.. Do not turn ths pge untl nstructed to do so. 3. Ths s closed book exmnton. You my hve your hnd clcultor

More information

LAPLACE TRANSFORM SOLUTION OF THE PROBLEM OF TIME-FRACTIONAL HEAT CONDUCTION IN A TWO-LAYERED SLAB

LAPLACE TRANSFORM SOLUTION OF THE PROBLEM OF TIME-FRACTIONAL HEAT CONDUCTION IN A TWO-LAYERED SLAB Journl of Appled Mthemtcs nd Computtonl Mechncs 5, 4(4), 5-3 www.mcm.pcz.pl p-issn 99-9965 DOI:.75/jmcm.5.4. e-issn 353-588 LAPLACE TRANSFORM SOLUTION OF THE PROBLEM OF TIME-FRACTIONAL HEAT CONDUCTION

More information

LOCAL FRACTIONAL LAPLACE SERIES EXPANSION METHOD FOR DIFFUSION EQUATION ARISING IN FRACTAL HEAT TRANSFER

LOCAL FRACTIONAL LAPLACE SERIES EXPANSION METHOD FOR DIFFUSION EQUATION ARISING IN FRACTAL HEAT TRANSFER Yn, S.-P.: Locl Frctonl Lplce Seres Expnson Method for Dffuson THERMAL SCIENCE, Yer 25, Vol. 9, Suppl., pp. S3-S35 S3 LOCAL FRACTIONAL LAPLACE SERIES EXPANSION METHOD FOR DIFFUSION EQUATION ARISING IN

More information

The Number of Rows which Equal Certain Row

The Number of Rows which Equal Certain Row Interntonl Journl of Algebr, Vol 5, 011, no 30, 1481-1488 he Number of Rows whch Equl Certn Row Ahmd Hbl Deprtment of mthemtcs Fcult of Scences Dmscus unverst Dmscus, Sr hblhmd1@gmlcom Abstrct Let be X

More information

2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification

2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification E395 - Pattern Recognton Solutons to Introducton to Pattern Recognton, Chapter : Bayesan pattern classfcaton Preface Ths document s a soluton manual for selected exercses from Introducton to Pattern Recognton

More information

Study of Trapezoidal Fuzzy Linear System of Equations S. M. Bargir 1, *, M. S. Bapat 2, J. D. Yadav 3 1

Study of Trapezoidal Fuzzy Linear System of Equations S. M. Bargir 1, *, M. S. Bapat 2, J. D. Yadav 3 1 mercn Interntonl Journl of Reserch n cence Technology Engneerng & Mthemtcs vlble onlne t http://wwwsrnet IN (Prnt: 38-349 IN (Onlne: 38-3580 IN (CD-ROM: 38-369 IJRTEM s refereed ndexed peer-revewed multdscplnry

More information

CIS587 - Artificial Intelligence. Uncertainty CIS587 - AI. KB for medical diagnosis. Example.

CIS587 - Artificial Intelligence. Uncertainty CIS587 - AI. KB for medical diagnosis. Example. CIS587 - rtfcl Intellgence Uncertnty K for medcl dgnoss. Exmple. We wnt to uld K system for the dgnoss of pneumon. rolem descrpton: Dsese: pneumon tent symptoms fndngs, l tests: Fever, Cough, leness, WC

More information

Lecture 12: Discrete Laplacian

Lecture 12: Discrete Laplacian Lecture 12: Dscrete Laplacan Scrbe: Tanye Lu Our goal s to come up wth a dscrete verson of Laplacan operator for trangulated surfaces, so that we can use t n practce to solve related problems We are mostly

More information

Demand. Demand and Comparative Statics. Graphically. Marshallian Demand. ECON 370: Microeconomic Theory Summer 2004 Rice University Stanley Gilbert

Demand. Demand and Comparative Statics. Graphically. Marshallian Demand. ECON 370: Microeconomic Theory Summer 2004 Rice University Stanley Gilbert Demnd Demnd nd Comrtve Sttcs ECON 370: Mcroeconomc Theory Summer 004 Rce Unversty Stnley Glbert Usng the tools we hve develoed u to ths ont, we cn now determne demnd for n ndvdul consumer We seek demnd

More information

Administrivia CSE 190: Reinforcement Learning: An Introduction

Administrivia CSE 190: Reinforcement Learning: An Introduction Administrivi CSE 190: Reinforcement Lerning: An Introduction Any emil sent to me bout the course should hve CSE 190 in the subject line! Chpter 4: Dynmic Progrmming Acknowledgment: A good number of these

More information

Online Appendix to. Mandating Behavioral Conformity in Social Groups with Conformist Members

Online Appendix to. Mandating Behavioral Conformity in Social Groups with Conformist Members Onlne Appendx to Mndtng Behvorl Conformty n Socl Groups wth Conformst Members Peter Grzl Andrze Bnk (Correspondng uthor) Deprtment of Economcs, The Wllms School, Wshngton nd Lee Unversty, Lexngton, 4450

More information

M/G/1/GD/ / System. ! Pollaczek-Khinchin (PK) Equation. ! Steady-state probabilities. ! Finding L, W q, W. ! π 0 = 1 ρ

M/G/1/GD/ / System. ! Pollaczek-Khinchin (PK) Equation. ! Steady-state probabilities. ! Finding L, W q, W. ! π 0 = 1 ρ M/G//GD/ / System! Pollcze-Khnchn (PK) Equton L q 2 2 λ σ s 2( + ρ ρ! Stedy-stte probbltes! π 0 ρ! Fndng L, q, ) 2 2 M/M/R/GD/K/K System! Drw the trnston dgrm! Derve the stedy-stte probbltes:! Fnd L,L

More information

NUMERICAL DIFFERENTIATION

NUMERICAL DIFFERENTIATION NUMERICAL DIFFERENTIATION 1 Introducton Dfferentaton s a method to compute the rate at whch a dependent output y changes wth respect to the change n the ndependent nput x. Ths rate of change s called the

More information

Lecture 4: November 17, Part 1 Single Buffer Management

Lecture 4: November 17, Part 1 Single Buffer Management Lecturer: Ad Rosén Algorthms for the anagement of Networs Fall 2003-2004 Lecture 4: November 7, 2003 Scrbe: Guy Grebla Part Sngle Buffer anagement In the prevous lecture we taled about the Combned Input

More information

INTERPOLATION(1) ELM1222 Numerical Analysis. ELM1222 Numerical Analysis Dr Muharrem Mercimek

INTERPOLATION(1) ELM1222 Numerical Analysis. ELM1222 Numerical Analysis Dr Muharrem Mercimek ELM Numercl Anlss Dr Muhrrem Mercmek INTEPOLATION ELM Numercl Anlss Some of the contents re dopted from Lurene V. Fusett, Appled Numercl Anlss usng MATLAB. Prentce Hll Inc., 999 ELM Numercl Anlss Dr Muhrrem

More information

Many-Body Calculations of the Isotope Shift

Many-Body Calculations of the Isotope Shift Mny-Body Clcultons of the Isotope Shft W. R. Johnson Mrch 11, 1 1 Introducton Atomc energy levels re commonly evluted ssumng tht the nucler mss s nfnte. In ths report, we consder correctons to tomc levels

More information

Bases for Vector Spaces

Bases for Vector Spaces Bses for Vector Spces 2-26-25 A set is independent if, roughly speking, there is no redundncy in the set: You cn t uild ny vector in the set s liner comintion of the others A set spns if you cn uild everything

More information

Optimal Resource Allocation and Policy Formulation in Loosely-Coupled Markov Decision Processes

Optimal Resource Allocation and Policy Formulation in Loosely-Coupled Markov Decision Processes Optml Resource Allocton nd Polcy Formulton n Loosely-Coupled Mrkov Decson Processes Dmtr A. Dolgov nd Edmund H. Durfee Deprtment of Electrcl Engneerng nd Computer Scence Unversty of Mchgn Ann Arbor, MI

More information

We consider a finite-state, finite-action, infinite-horizon, discounted reward Markov decision process and

We consider a finite-state, finite-action, infinite-horizon, discounted reward Markov decision process and MANAGEMENT SCIENCE Vol. 53, No. 2, Februry 2007, pp. 308 322 ssn 0025-1909 essn 1526-5501 07 5302 0308 nforms do 10.1287/mnsc.1060.0614 2007 INFORMS Bs nd Vrnce Approxmton n Vlue Functon Estmtes She Mnnor

More information

Smart Motorways HADECS 3 and what it means for your drivers

Smart Motorways HADECS 3 and what it means for your drivers Vehcle Rentl Smrt Motorwys HADECS 3 nd wht t mens for your drvers Vehcle Rentl Smrt Motorwys HADECS 3 nd wht t mens for your drvers You my hve seen some news rtcles bout the ntroducton of Hghwys Englnd

More information

SUMMER KNOWHOW STUDY AND LEARNING CENTRE

SUMMER KNOWHOW STUDY AND LEARNING CENTRE SUMMER KNOWHOW STUDY AND LEARNING CENTRE Indices & Logrithms 2 Contents Indices.2 Frctionl Indices.4 Logrithms 6 Exponentil equtions. Simplifying Surds 13 Opertions on Surds..16 Scientific Nottion..18

More information

A Family of Multivariate Abel Series Distributions. of Order k

A Family of Multivariate Abel Series Distributions. of Order k Appled Mthemtcl Scences, Vol. 2, 2008, no. 45, 2239-2246 A Fmly of Multvrte Abel Seres Dstrbutons of Order k Rupk Gupt & Kshore K. Ds 2 Fculty of Scence & Technology, The Icf Unversty, Agrtl, Trpur, Ind

More information

Linear and Nonlinear Optimization

Linear and Nonlinear Optimization Lner nd Nonlner Optmzton Ynyu Ye Deprtment of Mngement Scence nd Engneerng Stnford Unversty Stnford, CA 9430, U.S.A. http://www.stnford.edu/~yyye http://www.stnford.edu/clss/msnde/ Ynyu Ye, Stnford, MS&E

More information

The Regulated and Riemann Integrals

The Regulated and Riemann Integrals Chpter 1 The Regulted nd Riemnn Integrls 1.1 Introduction We will consider severl different pproches to defining the definite integrl f(x) dx of function f(x). These definitions will ll ssign the sme vlue

More information

Chapter 5 Supplemental Text Material R S T. ij i j ij ijk

Chapter 5 Supplemental Text Material R S T. ij i j ij ijk Chpter 5 Supplementl Text Mterl 5-. Expected Men Squres n the Two-fctor Fctorl Consder the two-fctor fxed effects model y = µ + τ + β + ( τβ) + ε k R S T =,,, =,,, k =,,, n gven s Equton (5-) n the textook.

More information

3/6/00. Reading Assignments. Outline. Hidden Markov Models: Explanation and Model Learning

3/6/00. Reading Assignments. Outline. Hidden Markov Models: Explanation and Model Learning 3/6/ Hdden Mrkov Models: Explnton nd Model Lernng Brn C. Wllms 6.4/6.43 Sesson 2 9/3/ courtesy of JPL copyrght Brn Wllms, 2 Brn C. Wllms, copyrght 2 Redng Assgnments AIMA (Russell nd Norvg) Ch 5.-.3, 2.3

More information

Simultaneous estimation of rewards and dynamics from noisy expert demonstrations

Simultaneous estimation of rewards and dynamics from noisy expert demonstrations Smultneous estmton of rewrds nd dynmcs from nosy expert demonstrtons Mchel Hermn,2, Tobs Gndele, Jo rg Wgner, Felx Schmtt, nd Wolfrm Burgrd2 - Robert Bosch GmbH - 70442 Stuttgrt - Germny 2- Unversty of

More information

7.2 Volume. A cross section is the shape we get when cutting straight through an object.

7.2 Volume. A cross section is the shape we get when cutting straight through an object. 7. Volume Let s revew the volume of smple sold, cylnder frst. Cylnder s volume=se re heght. As llustrted n Fgure (). Fgure ( nd (c) re specl cylnders. Fgure () s rght crculr cylnder. Fgure (c) s ox. A

More information

Abhilasha Classes Class- XII Date: SOLUTION (Chap - 9,10,12) MM 50 Mob no

Abhilasha Classes Class- XII Date: SOLUTION (Chap - 9,10,12) MM 50 Mob no hlsh Clsses Clss- XII Dte: 0- - SOLUTION Chp - 9,0, MM 50 Mo no-996 If nd re poston vets of nd B respetvel, fnd the poston vet of pont C n B produed suh tht C B vet r C B = where = hs length nd dreton

More information

The solution of transport problems by the method of structural optimization

The solution of transport problems by the method of structural optimization Scentfc Journls Mrtme Unversty of Szczecn Zeszyty Nukowe Akdem Morsk w Szczecne 2013, 34(106) pp. 59 64 2013, 34(106) s. 59 64 ISSN 1733-8670 The soluton of trnsport problems by the method of structurl

More information

Online Learning Algorithms for Stochastic Water-Filling

Online Learning Algorithms for Stochastic Water-Filling Onlne Lernng Algorthms for Stochstc Wter-Fllng Y G nd Bhskr Krshnmchr Mng Hseh Deprtment of Electrcl Engneerng Unversty of Southern Clforn Los Angeles, CA 90089, USA Eml: {yg, bkrshn}@usc.edu Abstrct Wter-fllng

More information

Online Stochastic Matching: New Algorithms with Better Bounds

Online Stochastic Matching: New Algorithms with Better Bounds Onlne Stochstc Mtchng: New Algorthms wth Better Bounds Ptrck Jllet Xn Lu My 202; revsed Jnury 203, June 203 Abstrct We consder vrnts of the onlne stochstc bprtte mtchng problem motvted by Internet dvertsng

More information

Soft Set Theoretic Approach for Dimensionality Reduction 1

Soft Set Theoretic Approach for Dimensionality Reduction 1 Interntonl Journl of Dtbse Theory nd pplcton Vol No June 00 Soft Set Theoretc pproch for Dmensonlty Reducton Tutut Herwn Rozd Ghzl Mustf Mt Ders Deprtment of Mthemtcs Educton nversts hmd Dhln Yogykrt Indones

More information

We partition C into n small arcs by forming a partition of [a, b] by picking s i as follows: a = s 0 < s 1 < < s n = b.

We partition C into n small arcs by forming a partition of [a, b] by picking s i as follows: a = s 0 < s 1 < < s n = b. Mth 255 - Vector lculus II Notes 4.2 Pth nd Line Integrls We begin with discussion of pth integrls (the book clls them sclr line integrls). We will do this for function of two vribles, but these ides cn

More information

Formulated Algorithm for Computing Dominant Eigenvalue. and the Corresponding Eigenvector

Formulated Algorithm for Computing Dominant Eigenvalue. and the Corresponding Eigenvector Int. J. Contemp. Mth. Scences Vol. 8 23 no. 9 899-9 HIKARI Ltd www.m-hkr.com http://dx.do.org/.2988/jcms.23.3674 Formulted Algorthm for Computng Domnnt Egenlue nd the Correspondng Egenector Igob Dod Knu

More information

Chapter 2 Introduction to Algebra. Dr. Chih-Peng Li ( 李 )

Chapter 2 Introduction to Algebra. Dr. Chih-Peng Li ( 李 ) Chpter Introducton to Algebr Dr. Chh-Peng L 李 Outlne Groups Felds Bnry Feld Arthetc Constructon of Glos Feld Bsc Propertes of Glos Feld Coputtons Usng Glos Feld Arthetc Vector Spces Groups 3 Let G be set

More information

More metrics on cartesian products

More metrics on cartesian products More metrcs on cartesan products If (X, d ) are metrc spaces for 1 n, then n Secton II4 of the lecture notes we defned three metrcs on X whose underlyng topologes are the product topology The purpose of

More information

Lecture 22: Logic Synthesis (1)

Lecture 22: Logic Synthesis (1) Lecture 22: Logc Synthess (1) Sldes courtesy o Demng Chen Some sldes Courtesy o Pro. J. Cong o UCLA Outlne Redng Synthess nd optmzton o dgtl crcuts, G. De Mchel, 1994, Secton 2.5-2.5.1 Overvew Boolen lgebr

More information

Research Article On the Upper Bounds of Eigenvalues for a Class of Systems of Ordinary Differential Equations with Higher Order

Research Article On the Upper Bounds of Eigenvalues for a Class of Systems of Ordinary Differential Equations with Higher Order Hndw Publshng Corporton Interntonl Journl of Dfferentl Equtons Volume 0, Artcle ID 7703, pges do:055/0/7703 Reserch Artcle On the Upper Bounds of Egenvlues for Clss of Systems of Ordnry Dfferentl Equtons

More information

Math 426: Probability Final Exam Practice

Math 426: Probability Final Exam Practice Mth 46: Probbility Finl Exm Prctice. Computtionl problems 4. Let T k (n) denote the number of prtitions of the set {,..., n} into k nonempty subsets, where k n. Argue tht T k (n) kt k (n ) + T k (n ) by

More information

For the percentage of full time students at RCC the symbols would be:

For the percentage of full time students at RCC the symbols would be: Mth 17/171 Chpter 7- ypothesis Testing with One Smple This chpter is s simple s the previous one, except it is more interesting In this chpter we will test clims concerning the sme prmeters tht we worked

More information

CENTROID (AĞIRLIK MERKEZİ )

CENTROID (AĞIRLIK MERKEZİ ) CENTOD (ĞLK MEKEZİ ) centrod s geometrcl concept rsng from prllel forces. Tus, onl prllel forces possess centrod. Centrod s tougt of s te pont were te wole wegt of pscl od or sstem of prtcles s lumped.

More information

Zbus 1.0 Introduction The Zbus is the inverse of the Ybus, i.e., (1) Since we know that

Zbus 1.0 Introduction The Zbus is the inverse of the Ybus, i.e., (1) Since we know that us. Introducton he us s the nverse of the us,.e., () Snce we now tht nd therefore then I V () V I () V I (4) So us reltes the nodl current njectons to the nodl voltges, s seen n (4). In developng the power

More information

Affine transformations and convexity

Affine transformations and convexity Affne transformatons and convexty The purpose of ths document s to prove some basc propertes of affne transformatons nvolvng convex sets. Here are a few onlne references for background nformaton: http://math.ucr.edu/

More information

princeton univ. F 17 cos 521: Advanced Algorithm Design Lecture 7: LP Duality Lecturer: Matt Weinberg

princeton univ. F 17 cos 521: Advanced Algorithm Design Lecture 7: LP Duality Lecturer: Matt Weinberg prnceton unv. F 17 cos 521: Advanced Algorthm Desgn Lecture 7: LP Dualty Lecturer: Matt Wenberg Scrbe: LP Dualty s an extremely useful tool for analyzng structural propertes of lnear programs. Whle there

More information

The Parametric Solution of Underdetermined Linear ODEs 1

The Parametric Solution of Underdetermined Linear ODEs 1 ISSN 036-7688, Progrmmng nd Computer Softwre, 0, Vol. 37, No., pp. 6 70. Pledes Publshng, Ltd., 0. The Prmetrc Soluton of Underdetermned Lner ODEs T. Wolf Deprtment of Mthemtcs, Brock Unversty, 500 Glenrdge

More information

p-adic Egyptian Fractions

p-adic Egyptian Fractions p-adic Egyptin Frctions Contents 1 Introduction 1 2 Trditionl Egyptin Frctions nd Greedy Algorithm 2 3 Set-up 3 4 p-greedy Algorithm 5 5 p-egyptin Trditionl 10 6 Conclusion 1 Introduction An Egyptin frction

More information

Dynamic Power Management in a Mobile Multimedia System with Guaranteed Quality-of-Service

Dynamic Power Management in a Mobile Multimedia System with Guaranteed Quality-of-Service Dynmc Power Mngement n Moble Multmed System wth Gurnteed Qulty-of-Servce Abstrct In ths pper we ddress the problem of dynmc power mngement n dstrbuted multmed system wth requred qulty of servce (QoS).

More information

Expected Value and Variance

Expected Value and Variance MATH 38 Expected Value and Varance Dr. Neal, WKU We now shall dscuss how to fnd the average and standard devaton of a random varable X. Expected Value Defnton. The expected value (or average value, or

More information

Feature Selection: Part 1

Feature Selection: Part 1 CSE 546: Machne Learnng Lecture 5 Feature Selecton: Part 1 Instructor: Sham Kakade 1 Regresson n the hgh dmensonal settng How do we learn when the number of features d s greater than the sample sze n?

More information