On Optimal Foraging and Multi-armed Bandits

Size: px
Start display at page:

Download "On Optimal Foraging and Multi-armed Bandits"

Transcription

1 Proceedngs of he 51s Annual Alleron Conference on Communcaon, Conrol and Compung, Ocober 013 On Opmal Foragng and Mul-armed Bands Vabhav Srvasava Paul Reverdy Naom E. Leonard Absrac We consder wo varans of he sandard mularmed band problem, namely, he mul-armed band problem wh ranson coss and he mul-armed band problem on graphs. We develop bloc allocaon algorhms for hese problems ha acheve an expeced cumulave regre ha s unformly domnaed by a logarhmc funcon of me, and an expeced cumulave number of ransons from one arm o anoher arm unformly domnaed by a double-logarhmc funcon of me. We observe ha he mul-armed band problem wh ranson coss and he assocaed bloc allocaon algorhm capure he ey feaures of popular anmal foragng models n leraure. I. INTRODUCTION Foragng s a fundamenal anmal behavor ha perans o searchng ou food resources and explong hem. Foragng behavor s suded n behavoral ecology usng economc prncples,.e., he foragng decsons are evaluaed based on her effecs on ceran pay-off funcons. A he hear of a foragng decson s he radeoff beween exploraon o search for a beer food resource) and exploaon o sc wh he bes nown food resource). In he engneerng leraure, a benchmar seup o sudy he exploraon-exploaon radeoff s he mul-armed band problem. The mul-armed band problem models a class of resource allocaon problems n whch a decson-maer allocaes a sngle resource by sequenally choosng one among a se of compeng alernave opons called arms. In he so-called saonary mul-armed band problem, a decson-maer a each dscree me nsan chooses an arm and collecs a reward drawn from an unnown saonary probably dsrbuon assocaed wh he seleced arm. The objecve of he decson-maer s o maxmze he oal reward aggregaed over he sequenal allocaon process. The fundamenal exploraon-exploaon radeoff n foragng can be modeled as a mul-armed band problem, and he effecveness of he foragng decsons can be measured by comparng hem o he opmal decsons for he mul-armed band problem. In hs paper, we explore hs connecon and argue ha he soluon o a Bayesan mularmed band problem capures he qualave feaures of he foragng behavor n some anmals. Leraure revew: The mul-armed band problem has been exensvely suded; a survey s presened n [1]. In her semnal wor, La and Robbns [] esablshed a logarhmc lower bound on he expeced number of mes a sub-opmal arm needs o be seleced by an opmal polcy. Ths research has been suppored n par by ONR gran N and ARO gran W911NG P. Reverdy s suppored hrough an NDSEG Fellowshp. V. Srvasava, P. Reverdy, and N. E. Leonard are wh he Deparmen of Mechancal and Aerospace Engneerng, Prnceon Unversy, Prnceon, NJ, USA {vabhavs, preverdy, Snce [], a consderable emphass has been on he desgn of smple heursc polces ha acheve he logarhmc lower bound on he expeced number of selecon nsances of any subopmal arm. To hs end, Auer e al. [3] developed upper confdence bound UCB) algorhms for mularmed bands wh bounded reward ha acheve logarhmc expeced cumulave regre unformly n me. Recenly, Srnvas e al. [4] developed asympocally opmal UCB algorhms for Gaussan process opmzaon. Kauffman e al. [5] developed a generc Bayesan UCB algorhm and esablshed s opmaly for bnary bands wh unform pror. Reverdy e al. [6] esablshed he opmaly of a Bayesan UCB algorhm for Gaussan rewards and drew several connecons beween hese algorhms and human decson-mang. They also elucdaed he role of prors n decson-mang performance. Some varaons of he mul-armed band problem have been suded as well. Agarwal e al. [7] suded he mularmed band problem wh ranson coss,.e., he mularmed band problem n whch a ceran penaly s mposed each me he decson-maer swches from he currenly seleced arm, and developed an asympocally opmal bloc allocaon algorhm. In hs paper, we consder he Gaussan mul-armed band problem wh ranson coss and develop a bloc allocaon algorhm ha acheves an expeced cumulave regre ha s unformly domnaed by a logarhmc erm. Moreover, he bloc allocaon scheme desgned n hs paper ncurs smaller expeced ranson coss han he bloc allocaon scheme n [7]. Klenberg e al. [8] consdered he mul-armed band problem n whch every arm s no avalable for selecon a each me sleepng expers), and hey analyzed he performance of he UCB algorhms. In conras o he emporal unavalably of arms n [8], we consder a spaal unavalably of arms. We propose a novel mul-armed band problem, namely, he graphcal mul-armed band problem, n whch only a subse of he arms can be seleced a he nex allocaon nsance gven he currenly seleced arm. We develop a bloc allocaon algorhm for such a problem ha acheves expeced cumulave regre ha s unformly domnaed by a logarhmc erm. Foragng has been exensvely suded n he behavoral ecology leraure [9], [10], [11], [1], [13]. A parcular emphass has been on opmal foragng heory [9], [10] ha sudes foragng behavor based on economc prncples. Tradonal wors [9], [10] n opmal foragng heory have suded he opmal behavor by ) pcng an approprae currency; ) esablshng approprae cos-benef funcons; and ) deermnng he opmal polces. Typcally he currency s chosen as he ne rae of energy nae and he fundamenal hypohess s ha hs nae rae s maxmzed.

2 The fundamenal quesons suded n opmal foragng heory nclude ) whch envronmen pach should he anmal vs nex? ) how long should he anmal say n ha pach? and ) whch foragng pah should he anmal choose n each pach? In recen years, a sgnfcan focus has been on he macroscopc properes of foragng. I has been observed ha Lévy flghs are effcen search mechansms, and has been hypoheszed ha anmal foragng has evolved no a Lévy flgh [14], [11]. An alernave macroscopc model o he Lévy flgh model s he nermen search model [15]. The nermen search model vews foragng n wo alernang phases. In he frs phase he anmal performs a local Brownan search, and n he second phase he anmal performs a ballsc relocaon. In boh he Lévy flgh and nermen search models, he ey macroscopc observaon s ha he anmal performs a local exploraon for some me and hen moves o a far-off locaon. Whle hese macroscopc models capure he general characerscs of foragng well, hey do no provde nsghs no he decson mechansms used by he anmal. There have been sgnfcan effors o undersand he decson mechansms n foragng; see, e.g., [16], [17], [18]. Of parcular neres here are he foragng sudes n he mul-armed band problem seng. Krebs e al. [16] suded foragng n grea-s n a wo-armed band seng and found ha he foragng polcy of grea-s s close o he opmal polcy for he wo-armed band problem. Keasar [17] explored he foragng behavor of bumblebees n a wo-armed band seng and dscussed plausble decson-mang mechansms. Conrbuons: In hs paper, we sudy he mul-armed band problem wh Gaussan rewards. In anmal foragng, he energy aggregaed from a pach can be hough of as he reward from he pach, and he anmal s objecve s o maxmze nae energy rae, whle mnmzng expendure n me and energy. In roboc foragng, he robo searches an area, and he reward s he aggregaed evdence. Analogous o he anmal, he robo s objecve s ypcally o maxmze evdence colleced, whle mnmzng expendure of me and energy. To address hs common problem, we consder wo parcular exensons of he sandard mul-armed band problem, namely, he mul-armed band problem wh ranson coss, and he graphcal mul-armed band problem. We jusfy he need for he exensons as follows. In he sandard mul-armed band problem, he decson-maer can swch beween wo arms any number of mes; whle n he roboc as well as he anmal foragng as, a hgher number of swches beween arms s undesrable because resuls n a hgher ravel me ha leads o a smaller energy/evdence aggregaon rae and a larger fuel cos. Thus he foragng objecve s equvalen o maxmzng he aggregaed reward whle mnmzng he swches beween he arms; hs s addressed by our frs exenson. Anoher shorcomng of he sandard mul-armed band problem s ha assumes ha each arm can be drecly vsed from anoher arm; whle hs s rue for any convex envronmen, non-convex envronmens requre exra care. A well nown echnque o handle nonconvex envronmens s he occupancy grd [19] ha consrucs a graph assocaed wh he non-convex envronmen. Accordngly, our second exenson o he mul-armed band problem on graphs enables sudy of he foragng problem n non-convex envronmens. The major conrbuons of hs wor are hreefold. Frs, we sudy he Gaussan mul-armed band problem wh ranson coss and exend he Bayesan-UCB algorhm n [6] o a bloc allocaon sraegy ha unformly acheves an expeced cumulave regre ha s domnaed by a logarhmc erm and an expeced number of ransons beween arms ha s domnaed by a double-logarhmc erm. Second, we sudy he graphcal Gaussan mul-armed band problem and exend he bloc allocaon sraegy o hs problem. We show ha even for he graphcal mul-armed band problem, he bloc allocaon sraegy unformly acheves an expeced cumulave regre ha s domnaed by a logarhmc erm. Thrd, we draw connecons beween anmal foragng behavor and he behavor of he proposed polces for he mul-armed bands. We argue ha he mul-armed bands and he assocaed bloc allocaon algorhms qualavely capure he foragng behavor of some anmals. In parcular, we observe ha he mul-armed band problem seup has he poenal o provde an overarchng framewor ha brngs ogeher he classcal opmal foragng heory, he Lévy flgh based macroscopc search models, and he decsonmechansm based search models. Paper srucure: The remander of he paper s organzed as follows. We revew sandard Gaussan mul-armed bands n Secon II. The Gaussan mul-armed bands wh ranson coss and he graphcal Gaussan mul-armed bands are suded n Secon III and IV, respecvely. We draw comparsons beween he behavor of he bloc allocaon algorhm and anmal foragng n Secon V and conclude n Secon VI. II. REVIEW OF BANDITS WITH GAUSSIAN REWARDS Consder an N-armed band problem,.e., a mul-armed band problem wh N arms. The reward assocaed wh arm {1,..., N} s a Gaussan random varable wh an unnown mean m, and a nown varance σs. The mean of he Gaussan reward a arm can be nerpreed as he sgnal srengh a he arm, whle he varance can be nerpreed as he samplng nose ha s he same a each arm. Le he agen choose arm a me {1,..., T } and receve a reward r N m, σs). The decson-maer s objecve s o choose a sequence of arms { } {1,...,T } ha maxmzes he expeced cumulave reward T m, where T s he horzon lengh of he sequenal allocaon process. For a mul-armed band, he expeced regre a me s defned by R = m m, where m = max{m {1,..., N}}. The objecve of he decson-maer can be equvalenly defned as mnmzng he expeced cumulave regre defned by T R = N =1 E[n T ], where nt s he cumulave number of mes opon has been chosen unl me T and = m m s he expeced regre due o pcng arm nsead of arm.

3 A. Bound on Opmal Performance La and Robbns [] showed ha any asympocally effcen algorhm for he mul-armed band problem mus choose subopmal arms for an expeced number of mes ha s a leas logarhmc n me. Tha ) s, E[n T 1 ] Dp p ) + o1) log T, where o1) 0 as T + and D ) R 0 {+ }, s defned by Dp p ) = p r) log p r) p r) dr, s he Kullbac-Lebler dvergence beween he reward densy p of any subopmal opon and he reward densy p of he opmal arm. For he Gaussan reward srucure consdered n hs paper, he Kullbac-Lebler dvergence s equal o Dp p ) = /σ s, and consequenly, E[n T ] σs/ + o1)) log T. Ths leads o a lower bound on he cumulave regre gven by σ ) R s + o1) log T. =1 B. Upper Credble Lm Algorhm for Gaussan Bands Le he pror on he mean reward a arm be a Gaussan random varable wh mean µ 0 and varance σ0. We are parcularly neresed n he case of an unnformave pror,.e., σ0 +. Le he number of mes arm has been chosen unl me be denoed by n. Le he emprcal mean of he rewards from arm unl me be m. Condoned on he number of vss n o arm and he emprcal mean m, he poseror dsrbuon of he mean reward M ) a arm a me s a Gaussan random varable wh mean and varance µ := E[M n, m ] = δ µ 0 + n m δ + n, and ) σ := Var[M n, m ] = σ s δ + n, respecvely, where δ = σs/σ 0. The UCL algorhm, proposed n [6], a each dscree) me frs compues he 1 1/K)-upper credble lm Q assocaed wh each arm {1,..., N} defned by Q := µ σ s + Φ ), δ + n K where K > 0 s a consan and Φ 1 ) s he nverse cumulave dsrbuon funcon for he sandard normal random varable. The UCL algorhm hen selecs an arm := arg max{q {1,..., N}}. For he unnformave pror,.e., δ 0 +, he UCL algorhm acheves a logarhmc expeced cumulave regre for a mul-armed band problem wh Gaussan rewards. In parcular, he regre sasfes he followng unform upper bound: R UCL =1 8β σ s + πe ) log T + 4β σs 1 log log log T ) ), πe where R UCL s he regre of he UCL algorhm a me, and β = 1.0. III. GAUSSIAN MULTI-ARMED BANDITS WITH TRANSITION COSTS Consder he N-armed band problem descrbed n Secon II. Suppose he decson-maer ncurs a random ranson cos c j R 0 for a ranson from arm o arm j. No cos s ncurred f he same arm as a he prevous me nsan s chosen,.e., c = 0. Such a cos srucure corresponds o a search problem n whch he N arms correspond o N spaally dsrbued regons and he ranson cos c j correspond o he ravel cos from regon o regon j. A. The Bloc UCL Algorhm For such Gaussan bands wh ranson coss, we develop a bloc allocaon sraegy ha exends he UCL algorhm of Secon II-B. To develop hs sraegy, we dvde he se of naural numbers allocaon nsances) no frames {f N} such ha frame f sars a me 1 and ends a me 1. Thus, he lengh of frame f s 1. We subdvde frame f no blocs ha we call rounds of allocaon. Le he frs 1 / blocs n frame f have lengh and he remanng allocaon nsances n frame f consue a sngle bloc of lengh 1 1 /. The oal number of allocaon rounds blocs) n frame f s b = 1 /. Le l N be he smalles ndex such ha T < l. Noe ha each round of allocaon s characerzed by he uple, r), for some {1,..., l}, and r {1,..., b }. The bloc UCL algorhm a each round of allocaon selecs he arm wh he maxmum upper lm o he smalles 1 1/Kτ r )-credble se Q r defned below), where τ r s he me a allocaon round, r), and chooses for he lengh of ha round bloc). B. Regre Analyss of he Bloc UCL Algorhm In hs secon, we analyze he regre of he bloc UCL algorhm. We frs nroduce some noaon. Le Q r be he maxmum upper lm o he smalles 1 1/Kτ r )-credble se for he mean of arm a allocaon round, r), where K = πe s he credble lm parameer. Le n r be he number of mes arm has been chosen unl allocaon round, r). Le s be he number of mes he decson-maer ransons o arm from anoher arm j {1,..., N} \ {} unl me. Le he emprcal mean of he rewards from arm unl allocaon round, r) be m r. Condoned on he number of vss n r o arm and he emprcal mean m r, he poseror dsrbuon of he mean reward M ) a arm a allocaon round, r) s a Gaussan random varable wh mean and varance µ r σ r := E[M n r, m r := Var[M n r, m r ] = respecvely. Moreover, E[µ r n r ]= δ µ 0 + nr m δ + n r ] = δ µ 0 + nr δ + n r σ s δ + n r and Var[µ r n r, m r, and ]= nr σs δ + n r ).

4 Accordngly, he 1 1 Kτ r )-upper credble lm Q r s Q r = µ r + σ s δ + n r Φ Kτ r ). Also, for each {1,..., N}, we defne consans γ 1 = 8β σ s + 1 log + K, γ = 4β σs 1 log ) K + log 4 K, γ3 = γ1 log log log ) 4β σ sγ1 log log γ c max = max{e[c j ] j {1,..., N}}. )1 + π 6 ), and Le {R BUCL } {1,...,T } be he sequence of he expeced regre of he bloc UCL algorhm, and {S BUCL } {1,...,T } be he sequence of expeced ranson coss. The Bloc UCL algorhm acheves a logarhmc expeced cumulave regre as formalzed n he followng heorem. Theorem 1 Regre of Bloc UCL Algorhm): The followng saemens hold for he Gaussan mul-armed band problem wh ranson coss and he bloc UCL algorhm wh an unnformave pror: ) he expeced number of mes a subopmal arm s chosen unl me T sasfes E[n T ] γ 1 log T 4β σ s log log T + γ ; ) he expeced number of ransons o a subopmal arm from anoher arm unl me T sasfes E[s T ] γ 1 log ) log log T + γ 3; ) he cumulave regre and he cumulave ranson cos unl me T sasfy R BUCL S BUCL =1 γ1 log T 4β σs ) log log T + γ, c max =1, + c max ) γ 1 log ) log log T + γ 3) + c max. Proof: See Appendx. IV. GRAPHICAL GAUSSIAN BANDITS We now consder mul-armed bands wh Gaussan rewards n whch he decson-maer canno move o every oher arm from he curren arm. Le he arms ha can be vsed from arm be ne) {1,..., N}. Such a mularmed band can be represened by a graph G wh node se {1,..., N} and edge se E = {, j) j ne), {1,..., N}}. We assume ha he graph s conneced n he sense ha here exss a leas one pah from each node {1,..., N} o every oher node j {1,..., N}. A. The Graphcal Bloc UCL Algorhm For he graphcal Gaussan bands, we develop an algorhm smlar o he bloc allocaon algorhm, namely, he graphcal bloc UCL algorhm. Smlar o he bloc allocaon algorhm, a each comparson bloc, he arm wh he maxmum upper credble lm s deermned. Snce he arm wh he maxmum upper credble lm may no be mmedaely reached from he curren arm, he graphcal bloc UCL algorhm raverses a shores pah from he curren arm o he arm wh he maxmum upper credble lm. The ey nuon behnd he algorhm s ha he bloc allocaon sraegy resuls n an expeced number of ransons ha s sub-logarhmc n he horzon lengh. In he conex of graphcal bands, sub-logarhmc ransons resul n sub-logarhmc undesred vss o he arms on he chosen shores pah o he desred arm. Consequenly, he regre of he algorhm s domnaed by he logarhmc erm. B. Regre Analyss of he Graphcal Bloc UCL Algorhm We now analyze he performance of he graphcal bloc UCL algorhm. Le {R GUCL } {1,...,T } be he sequence of expeced regre of he graphcal bloc UCL algorhm. The graphcal bloc UCL algorhm acheves a logarhmc expeced cumulave regre as formalzed n he followng heorem. Theorem Regre of Graphcal Bloc UCL Algorhm): The followng saemens hold for he graphcal Gaussan mul-armed band problem wh he graphcal bloc UCL algorhm and an unnformave pror: ) he expeced number of mes a subopmal arm s chosen unl me T sasfes E[n T ] γ1 log T 4β σs log log T + γ + =1, γ 1 log ) log log T + γ3 ) + 1; ) he cumulave regre unl me T sasfes +γ + R GUCL =1, Proof: See Appendx. γ1 log T 4β σs log log T =1 ) γ 1 log ) log log T +γ3) +1 ; V. COMPARISON WITH ANIMAL FORAGING In hs secon, we compare he behavor of he bloc allocaon algorhm for he mul-armed bands wh he anmal foragng behavor repored n he leraure. Consder he foragng envronmen as composed of paches and each pach has sources of energy ha are modeled by Gaussan random varables wh an unnown mean and a nown varance. The exploraon-exploaon radeoff n he foragng problem can be modeled by he mul-armed band problem. In parcular, he foragng objecve of anmals s o maxmze he ne energy accumulaon rae whch n he mul-armed band seng maps o maxmzng he expeced cumulave

5 reward whle mnmzng he ravel me,.e., mnmzng he number of ransons among arms. The soluon o he mul-armed band problem naurally answers he frs wo fundamenal quesons suded n opmal foragng heory: ) whch envronmen pach should he anmal vs nex? ) how long should he anmal say n ha pach? Alhough he soluon o he mul-armed band problem does no answer he hrd fundamenal queson: whch foragng pah should he anmal choose n each pach? To undersand he hrd queson, s naural o envson ha pons whn a pach are hghly correlaed n erms of he energy accumulaon,.e., each pon whn a pach provdes energy a somewha he same rae, and accordngly he energy can be accumulaed, e.g., va an ergodc random wal. For smplcy of analyss, n hs paper, we assume ha he arms are uncorrelaed and he pror s unnformave. In general, he pror may be nformave and arms may be correlaed. The algorhm proposed n hs paper exends o hs case by smply replacng he N unvarae nference procedures wh an N-varae nference procedure. The correlaon srucure capures he srucure of he envronmen: hgher correlaon descrbes a smooher envronmen, whle lower correlaon descrbes a rougher envronmen. In a suffcenly correlaed envronmen, he bloc allocaon algorhm a allocaon round, r) pcs an arm wh hghes value of Q r and samples mes. A he subsequen allocaon nsance, due o he correlaon srucure he uncerany n he esmaes for he nearby locaons wll go down whle he uncerany n he far-off locaons would reman hgh. Consequenly, he componen of Q r assocaed wh he wdh of he credble se wll be hgher for he far-off locaons han he nearby locaons. If he pror means are assumed o be unform, he bloc allocaon sraegy a he nex allocaon nsance wll selec a locaon far-off from he curren locaon. Ths s a cenral feaure of he macroscopc foragng models, ncludng he Lévy flgh model and he nermen search model. Thus, he Bayesan mul-armed band problem and he assocaed bloc allocaon sraegy qualavely capures he behavor of Lévy flghs and relaed macroscopc models for search. Overall, he mul-armed band problem wh ranson coss models he fundamenal foragng objecve as defned n he opmal foragng leraure, and s soluon yelds search rajecores an o hose descrbed by macroscopc search models. Moreover, he soluon o he mul-armed band problem wh ranson coss naurally provdes he decson mechansms nvolved wh he search process. Therefore, he mul-armed band problem seup has he poenal o provde an overarchng framewor ha brngs ogeher he classcal opmal foragng heory, he Lévy flgh based macroscopc search models, and he decson-mechansm based search models. VI. CONCLUSIONS We suded wo varaons of he Gaussan mul-armed band problem, namely, he Gaussan mul-armed band problem wh ranson cos, and he graphcal Gaussan mul-armed band problem and developed bloc allocaon algorhms ha unformly acheve an expeced cumulave regre domnaed by a logarhmc funcon of me, and a number of expeced cumulave ransons among he arms domnaed by a double-logarhmc funcon of me. We drew some qualave connecons beween foragng behavor of some anmals and he behavor of he bloc allocaon algorhm. In parcular, we argued ha he mul-armed band problem models he foragng objecve n opmal foragng heory well and he assocaed bloc allocaon sraegy capures he ey feaures of popular macroscopc search models. A hs sage, we observe and pon ou he poenal of he mul-armed band problem and he assocaed bloc allocaon algorhm o brdge he gap beween classcal opmal foragng heory and recen macroscopc search models. Ths suggess an excng new avenue of nqury n whch he band model may prove valuable for fuure sudy of anmal foragng. In he fuure, we plan o nvesgae he band model more exensvely n he conex of emprcal wor on boh anmal and roboc foragng. APPENDIX A. Proof of regre of he bloc UCL algorhm Proof of Theorem 1: We sar by esablshng he frs saemen. For a gven, le, r ) be he lexcographcally maxmum uple such ha τ r. We noe ha n T = 1 = ) = η + l + η + l + 1 = & n r l r=1 < η) + 1 = & n r η) ) 1 = & n r η) b 1 τr = & n r η). 1) I can be shown see [0] for deals) ha f we choose η = 8β σ s log T 1 log log T ) + 4β σ s 1 log ), hen E[n T ] η + l + K l b r=1 τ r. ) We now focus on he erm l b r=1 τ r. We noe ha τ r = 1 + r 1), and hence b r=1 τ r = b r= r 1) b 1 dx x 1) log. 3) 1 Snce T l 1, follows ha l 1 + log T =: l. Therefore, nequales ) and 3) yeld

6 E[n T ] η + l + K l η + l + 8 K + log K l 1 + log ) γ1 log T 4β σs log log T + γ. We now esablsh he second saemen. In he spr of [7], we noe ha he number of mes he decson-maer ransons o arm from anoher arm n frame f s equal o he number of mes arm s seleced n frame dvded by he lengh of each bloc s frame f. Consequenly, s T l = nl l n + l 1 n 1 Therefore, follows ha = l n l 1 n n 1 ) nl + 1 l ] E[s T ] E[nl l l 1 + l 1 + n. E[n ]. 4) Subsung E[n ] n nequaly 4) wh he derved upper bounds and performng some algebrac manpulaons, we oban E[s T ] γ 1 log ) log log T + γ 3. We now esablsh he las saemen. The bound of he cumulave regre follows from he defnon and he frs saemen. To esablsh he bound on he cumulave swchng cos, we noe ha S BUCL c max E[s T ] + c max E[sT ] =1, c max =1, + c max )E[sT ] + c max, 5) where he second nequaly follows from he observaon ha s T T =1, st + 1. The fnal expresson follows from nequaly 5) and he second saemen. B. Proof of regre of he graphcal bloc UCL algorhm Proof of Theorem : We sar by esablshng he frs saemen. We classfy he selecon of arms n wo caegores, namely, he goal selecon and he ransen selecon. The goal selecon of an arm corresponds o he suaon n whch he arm has he maxmum upper credble lm, whle he ransen selecon corresponds o he suaon n whch he arm s seleced because belongs o he chosen shores pah o he arm wh he maxmum credble lm. We noe ha due o ransen selecons, he number of frames unl me T are a mos equal o he number of frames f here are no ransen selecons. Consequenly, he expeced number of goal selecons of a subopmal arm are upper bounded by he expeced number of selecons of arm n he bloc allocaon algorhm,.e., E[n T goal,] γ 1 log T 4β σ s log log T + γ. Moreover, he number of ransen selecons of arm are upper bounded by he oal number of ransons from an arm o anoher arm n he bloc allocaon algorhm,.e., E[n T ransen,] =1, γ 1 log ) log log T + γ3 ) + 1. The expeced number of selecons of arm s he sum of he expeced number of ransen selecons and he expeced number of goal selecons, and hus he frs saemen follows. The second saemen follows mmedaely from he defnon of he cumulave regre. REFERENCES [1] S. Bubec and N. Cesa-Banch. Regre analyss of sochasc and nonsochasc mul-armed band problems. Machne Learnng, 51):1 1, 01. [] T. L. La and H. Robbns. Asympocally effcen adapve allocaon rules. Advances n Appled Mahemacs, 61):4, [3] P. Auer, N. Cesa-Banch, and P. Fscher. Fne-me analyss of he mularmed band problem. Machne learnng, 47):35 56, 00. [4] N. Srnvas, A. Krause, S. M. Kaade, and M. Seeger. Informaonheorec regre bounds for Gaussan process opmzaon n he band seng. IEEE Transacons on Informaon Theory, 585): , 01. [5] E. Kaufmann, O. Cappé, and A. Garver. On Bayesan upper confdence bounds for band problems. In In. Conf. on Arfcal Inellgence and Sascs, pages , La Palma, Canary Islands, Span, Aprl 01. [6] P. Reverdy, V. Srvasava, and N. E. Leonard. Modelng human decson-mang n mul-armed bands. In Muldscplnary Conf. on Renforcemen Learnng and Decson Mang, Prnceon, NJ, USA, Oc 013. [7] R. Agrawal, M. V. Hedge, and D. Teneezs. Asympocally effcen adapve allocaon rules for he mul-armed band problem wh swchng cos. IEEE Transacons on Auomac Conrol, 3310): , [8] R. Klenberg, A. Nculescu-Mzl, and Y. Sharma. Regre bounds for sleepng expers and bands. Machne learnng, 80-3):45 7, 010. [9] G. H. Pye, H. R. Pullam, and E. L. Charnov. Opmal foragng: a selecve revew of heory and ess. Quarerly Revew of Bology, pages , [10] D. W. Sephens,, and J. R. Krebs. Foragng heory. Prnceon Unversy Press, [11] G. M. Vswanahan, M. G. E. da Luz, E. P. Raposo, and H. E. Sanley. The Physcs of Foragng: An Inroducon o Random Searches and Bologcal Encouners. Cambrdge Unversy Press, 011. [1] E. Gelenbe, N. Schmaju, J. Saddon, and J. Ref. Auonomous search by robos and anmals: A survey. Robocs and Auonomous Sysems, 1):3 34, [13] S. R. X. Dall, L. Graldeau, O. Olsson, J. M. McNamara, and D. W. Sephens. Informaon and s use by anmals n evoluonary ecology. Trends n Ecology & Evoluon, 04): , 005. [14] G. M. Vswanahan, S. V. Buldyrev, S. Havln, M. G. E. da Luz, E. P. Raposo, and H. E. Sanley. Opmzng he success of random searches. Naure, ): , [15] O. Bénchou, C. Loverdo, M. Moreau, and R. Vourez. Inermen search sraeges. Revews of Modern Physcs, 831):81, 011. [16] J. R. Krebs, A. Kaceln, and P. Taylor. Tes of opmal samplng by foragng grea s. Naure, ):7 31, [17] T. Keasar, E. Rashovch, D. Cohen, and A. Shmda. Bees n wo-armed band suaons: Foragng choces and possble decson mechansms. Behavoral Ecology, 136): , 00. [18] A. M. Hen and S. A. McKnley. Sensng and decson-mang n random search. Proceedngs of he Naonal Academy of Scences, 10930): , 01. [19] S. Thrun, W. Burgard, and D. Fox. Probablsc Robocs. The MIT Press, 005. [0] P. Reverdy, V. Srvasava, and N. E. Leonard. Modelng human decson-mang n generalzed Gaussan mul-armed bands. arxv preprn arxv: , July 013.

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

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

More information

arxiv: v3 [cs.lg] 14 Feb 2014

arxiv: v3 [cs.lg] 14 Feb 2014 1 Modelng Human Decson-mang n Generalzed Gaussan Mul-armed Bands Paul Reverdy Vabhav Srvasava Naom Ehrch Leonard arxv:1307.6134v3 [cs.lg] 14 Feb 014 Absrac We presen a formal model of human decsonmang

More information

Satisficing in Gaussian bandit problems

Satisficing in Gaussian bandit problems Sasfcng n Gaussan band problems Paul Reverdy and Naom E. Leonard Absrac We propose a sasfcng objecve for he mularmed band problem,.e., where he objecve s o acheve performance above a gven hreshold. We

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

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

( 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

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

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

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

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

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

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

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

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

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

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

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

More information

Sequential Sensor Selection and Access Decision for Spectrum Sharing

Sequential Sensor Selection and Access Decision for Spectrum Sharing Sequenal Sensor Selecon and Access Decson for Specrum Sharng Jhyun Lee, Suden Member, IEEE and Eylem Ekc, Fellow, IEEE Absrac We develop an algorhm for sequenal sensor selecon and channel access decson

More information

Sampling Procedure of the Sum of two Binary Markov Process Realizations

Sampling Procedure of the Sum of two Binary Markov Process Realizations Samplng Procedure of he Sum of wo Bnary Markov Process Realzaons YURY GORITSKIY Dep. of Mahemacal Modelng of Moscow Power Insue (Techncal Unversy), Moscow, RUSSIA, E-mal: gorsky@yandex.ru VLADIMIR KAZAKOV

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

Appendix to Online Clustering with Experts

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

More information

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

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

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

Comparison of Differences between Power Means 1

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

More information

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

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

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

Optimal environmental charges under imperfect compliance

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

More information

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

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

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

2 Aggregate demand in partial equilibrium static framework

2 Aggregate demand in partial equilibrium static framework Unversy of Mnnesoa 8107 Macroeconomc Theory, Sprng 2009, Mn 1 Fabrzo Perr Lecure 1. Aggregaon 1 Inroducon Probably so far n he macro sequence you have deal drecly wh represenave consumers and represenave

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

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

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

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

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

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

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

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

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

1 Stochastic Multi-armed Bandit Problems. 3 Animal Foraging and Multi-armed Bandit Problems

1 Stochastic Multi-armed Bandit Problems. 3 Animal Foraging and Multi-armed Bandit Problems Oulne The Sochasc Mul-Armed Band Problem: In Neuroscence, Ecology, and Engneerng Vabhav Srvasava CYber Physcal Human sysems Research (CYPHER) Lab Deparmen of Elecrcal & Compuer Engneerng Mchgan Sae Unversy

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

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

[ ] 2. [ ]3 + (Δx i + Δx i 1 ) / 2. Δx i-1 Δx i Δx i+1. TPG4160 Reservoir Simulation 2018 Lecture note 3. page 1 of 5

[ ] 2. [ ]3 + (Δx i + Δx i 1 ) / 2. Δx i-1 Δx i Δx i+1. TPG4160 Reservoir Simulation 2018 Lecture note 3. page 1 of 5 TPG460 Reservor Smulaon 08 page of 5 DISCRETIZATIO OF THE FOW EQUATIOS As we already have seen, fne dfference appromaons of he paral dervaves appearng n he flow equaons may be obaned from Taylor seres

More information

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

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

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

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

More information

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

An Optimal Control Approach to the Multi-agent Persistent Monitoring Problem

An Optimal Control Approach to the Multi-agent Persistent Monitoring Problem An Opmal Conrol Approach o he Mul-agen Perssen Monorng Problem Chrsos.G. Cassandras, Xuchao Ln and Xu Chu Dng Dvson of Sysems Engneerng and Cener for Informaon and Sysems Engneerng Boson Unversy, cgc@bu.edu,

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

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

Computing Relevance, Similarity: The Vector Space Model

Computing Relevance, Similarity: The Vector Space Model Compung Relevance, Smlary: The Vecor Space Model Based on Larson and Hears s sldes a UC-Bereley hp://.sms.bereley.edu/courses/s0/f00/ aabase Managemen Sysems, R. Ramarshnan ocumen Vecors v ocumens are

More information

Bayesian Inference of the GARCH model with Rational Errors

Bayesian Inference of the GARCH model with Rational Errors 0 Inernaonal Conference on Economcs, Busness and Markeng Managemen IPEDR vol.9 (0) (0) IACSIT Press, Sngapore Bayesan Inference of he GARCH model wh Raonal Errors Tesuya Takash + and Tng Tng Chen Hroshma

More information

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

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

More information

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

( ) [ ] MAP Decision Rule

( ) [ ] MAP Decision Rule Announcemens Bayes Decson Theory wh Normal Dsrbuons HW0 due oday HW o be assgned soon Proec descrpon posed Bomercs CSE 90 Lecure 4 CSE90, Sprng 04 CSE90, Sprng 04 Key Probables 4 ω class label X feaure

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

Sequential Decision Making in Two-Dimensional Hypothesis Testing

Sequential Decision Making in Two-Dimensional Hypothesis Testing 5nd IEEE Conference on Decson and Conrol December 1-13, 13. Florence, Ialy Sequenal Decson Mang n Two-Dmensonal Hypohess Tesng Mchael Carlsle, Olympa Hadjlads Absrac In hs wor, we consder he problem of

More information

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

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

More information

arxiv: v2 [cs.lg] 22 Nov 2016

arxiv: v2 [cs.lg] 22 Nov 2016 Unmodal Thompson Samplng for Graph Srucured Arms Sefano Paladno and Francesco Trovò and Marcello Resell and Ncola Ga Dparmeno d Eleronca, Informazone e Bongegnera Polecnco d Mlano, Mlano, Ialy {sefano.paladno,

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

Multi-priority Online Scheduling with Cancellations

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

More information

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

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

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

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

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

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

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

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

Sampling Coordination of Business Surveys Conducted by Insee

Sampling Coordination of Business Surveys Conducted by Insee Samplng Coordnaon of Busness Surveys Conduced by Insee Faben Guggemos 1, Olver Sauory 1 1 Insee, Busness Sascs Drecorae 18 boulevard Adolphe Pnard, 75675 Pars cedex 14, France Absrac The mehod presenly

More information

Survival Analysis and Reliability. A Note on the Mean Residual Life Function of a Parallel System

Survival Analysis and Reliability. A Note on the Mean Residual Life Function of a Parallel System Communcaons n Sascs Theory and Mehods, 34: 475 484, 2005 Copyrgh Taylor & Francs, Inc. ISSN: 0361-0926 prn/1532-415x onlne DOI: 10.1081/STA-200047430 Survval Analyss and Relably A Noe on he Mean Resdual

More information

5th International Conference on Advanced Design and Manufacturing Engineering (ICADME 2015)

5th International Conference on Advanced Design and Manufacturing Engineering (ICADME 2015) 5h Inernaonal onference on Advanced Desgn and Manufacurng Engneerng (IADME 5 The Falure Rae Expermenal Sudy of Specal N Machne Tool hunshan He, a, *, La Pan,b and Bng Hu 3,c,,3 ollege of Mechancal and

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

A GENERAL FRAMEWORK FOR CONTINUOUS TIME POWER CONTROL IN TIME VARYING LONG TERM FADING WIRELESS NETWORKS

A GENERAL FRAMEWORK FOR CONTINUOUS TIME POWER CONTROL IN TIME VARYING LONG TERM FADING WIRELESS NETWORKS A GENERAL FRAEWORK FOR CONTINUOUS TIE POWER CONTROL IN TIE VARYING LONG TER FADING WIRELESS NETWORKS ohammed. Olama, Seddk. Djouad Charalambos D. Charalambous Elecrcal and Compuer Engneerng Deparmen Elecrcal

More information

WITH the proliferation of smart wireless devices and mobile

WITH the proliferation of smart wireless devices and mobile Ths arcle has been acceped for publcaon n a fuure ssue of hs journal, bu has no been fully eded Conen may change pror o fnal publcaon Caon nformaon: DOI 1119/TMC18847337, I Transacons on Moble Compung

More information

Deepanshu Vasal. Abstract. We consider a general finite-horizon non zero-sum dynamic game with asymmetric information with N selfish

Deepanshu Vasal. Abstract. We consider a general finite-horizon non zero-sum dynamic game with asymmetric information with N selfish Sequenal decomposon of dynamc games wh 1 asymmerc nformaon and dependen saes Deepanshu Vasal Absrac We consder a general fne-horzon non zero-sum dynamc game wh asymmerc nformaon wh N selfsh players, where

More information

2. SPATIALLY LAGGED DEPENDENT VARIABLES

2. SPATIALLY LAGGED DEPENDENT VARIABLES 2. SPATIALLY LAGGED DEPENDENT VARIABLES In hs chaper, we descrbe a sascal model ha ncorporaes spaal dependence explcly by addng a spaally lagged dependen varable y on he rgh-hand sde of he regresson equaon.

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

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

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

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

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

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

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

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

EXECUTION COSTS IN FINANCIAL MARKETS WITH SEVERAL INSTITUTIONAL INVESTORS

EXECUTION COSTS IN FINANCIAL MARKETS WITH SEVERAL INSTITUTIONAL INVESTORS EXECUION COSS IN FINANCIAL MARKES WIH SEVERAL INSIUIONAL INVESORS Somayeh Moazen, Yuyng L, Kae Larson Cheron School of Compuer Scence Unversy of Waerloo, Waerloo, ON, Canada emal: {smoazen, yuyng, klarson}@uwaerlooca

More information

F-Tests and Analysis of Variance (ANOVA) in the Simple Linear Regression Model. 1. Introduction

F-Tests and Analysis of Variance (ANOVA) in the Simple Linear Regression Model. 1. Introduction ECOOMICS 35* -- OTE 9 ECO 35* -- OTE 9 F-Tess and Analyss of Varance (AOVA n he Smple Lnear Regresson Model Inroducon The smple lnear regresson model s gven by he followng populaon regresson equaon, or

More information

SELFSIMILAR PROCESSES WITH STATIONARY INCREMENTS IN THE SECOND WIENER CHAOS

SELFSIMILAR PROCESSES WITH STATIONARY INCREMENTS IN THE SECOND WIENER CHAOS POBABILITY AD MATEMATICAL STATISTICS Vol., Fasc., pp. SELFSIMILA POCESSES WIT STATIOAY ICEMETS I TE SECOD WIEE CAOS BY M. M A E J I M A YOKOAMA AD C. A. T U D O LILLE Absrac. We sudy selfsmlar processes

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

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

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

Bandlimited channel. Intersymbol interference (ISI) This non-ideal communication channel is also called dispersive channel

Bandlimited channel. Intersymbol interference (ISI) This non-ideal communication channel is also called dispersive channel Inersymol nererence ISI ISI s a sgnal-dependen orm o nererence ha arses ecause o devaons n he requency response o a channel rom he deal channel. Example: Bandlmed channel Tme Doman Bandlmed channel Frequency

More information

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

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

More information

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

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

More information

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

Normal Random Variable and its discriminant functions

Normal Random Variable and its discriminant functions Noral Rando Varable and s dscrnan funcons Oulne Noral Rando Varable Properes Dscrnan funcons Why Noral Rando Varables? Analycally racable Works well when observaon coes for a corruped snle prooype 3 The

More information

SUPPLEMENT TO EFFICIENT DYNAMIC MECHANISMS IN ENVIRONMENTS WITH INTERDEPENDENT VALUATIONS: THE ROLE OF CONTINGENT TRANSFERS

SUPPLEMENT TO EFFICIENT DYNAMIC MECHANISMS IN ENVIRONMENTS WITH INTERDEPENDENT VALUATIONS: THE ROLE OF CONTINGENT TRANSFERS SUPPLEMENT TO EFFICIENT DYNAMIC MECHANISMS IN ENVIRONMENTS WITH INTERDEPENDENT VALUATIONS: THE ROLE OF CONTINGENT TRANSFERS HENG LIU In hs onlne appendx, we dscuss budge balance and surplus exracon n dynamc

More information