Satisficing in Gaussian bandit problems

Size: px
Start display at page:

Download "Satisficing in Gaussian bandit problems"

Transcription

1 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 show ha hs new problem s equvalen o a sandard mul-armed band problem wh a maxmzng objecve and use hs equvalence o fnd bounds on performance n erms of he sasfcng objecve. For he specal case of Gaussan rewards we show ha he sasfcng problem s equvalen o a relaed sandard mularmed band problem agan wh Gaussan rewards. We apply he Upper Credble Lm (UCL algorhm o hs sandard problem and show how acheves opmal performance n erms of he sasfcng objecve. I. INTRODUCTION Engneerng soluons o decson-makng problems are ofen desgned o maxmze an objecve funcon. However, n many conexs maxmzaon of an objecve funcon s an unreasonable goal, eher because he objecve self s poorly defned or because solvng he resulng opmzaon problem s nracable or cosly. In hese conexs, s valuable o consder alernave decson-makng frameworks. Herber Smon consdered [16 alernave models of raonal decson makng wh he goal of makng hem compable wh he access o nformaon and he compuaonal capaces ha are acually possessed by organsms, ncludng man, n he knds of envronmens n whch such organsms exs. A major feaure of he models he consdered s wha he called sasfcng. In [16, Smon dscussed n very broad erms a varey of smplfcaons o he classcal economc concep of raonaly, mos mporanly he dea ha payoffs should be smple, defned by dong well relave o some hreshold value. In [17, he nroduced he word sasfcng o refer o hs hresholdng concep and consdered an ecologcal example of food foragng behavor n deal usng mahemacal erms. He also brefly dscussed how sasfcng relaes o problems n nvenory conrol and more complcaed decson processes lke playng chess. Snce Smon s poneerng work, sasfcng has been suded n many felds such as psychology [15, economcs [3, managemen scence [10, [1, and ecology [0, [4. In engneerng, sasfcng s of neres for he same reasons ha movaed s nroducon n he socal scence leraure, specfcally ha can smplfy decson-makng problems. Furhermore, many engneerng problems are naurally posed usng a sasfcng objecve, for example desgn problems Ths research has been suppored n par by ONR grans N and N and ARO gran W911NG P. Reverdy s suppored hrough a NDSEG Fellowshp. The auhors are wh he Deparmen of Mechancal and Aerospace Engneerng, Prnceon Unversy, Prnceon, NJ 08544, USA {preverdy,naom}@prnceon.edu ha have o mee gven specfcaons. A desgn ha mees all he requred specfcaons s accepable, and he desgners may be ndfferen beween any such desgn. In hs conex, opmzaon may be poorly defned, for example f here are several compeng performance measures ha rade off n complcaed ways. Sasfcng can be a smpler decson paradgm han maxmzng, whch requres addonal nformaon abou preferences among possble radeoffs. Sasfcng has been suded n he engneerng leraure n several conexs. In [11, Nakayama suded desgn opmzaon usng a sasfcng objecve and found ha s effecve n many praccal felds. In [6, he auhors suded conrol heory usng a sasfcng objecve funcon, and n [, he auhors used sasfcng o sudy opmal sofware desgn. Sasfcng can be mplemened n a varey of ways. In hs paper, we consder he sochasc mul-armed band problem [14, where a decson maker sequenally chooses one of a se of alernave opons, or arms, and earns a reward drawn from a saonary probably dsrbuon assocaed wh ha arm. The sandard mul-armed band problem uses a maxmzng objecve, for whch here s a known performance bound. We propose a sasfcng objecve for he mularmed band problem based on he number of mes he decson maker receves a reward ha s above a hreshold value and show ha he mul-armed band problem wh hs objecve s equvalen o a relaed sandard mul-armed band problem. We use he equvalen problem o derve a performance bound for he new sasfcng problem. For Gaussan band problems,.e., where he reward dsrbuons are Gaussan wh unknown mean and known varance, we show ha solvng he problem wh he sasfcng objecve s equvalen o solvng a sandard Gaussan mularmed band problem. We hen apply he UCL algorhm we developed n prevous work [13 o he sandard problem, and show how hs algorhm acheves opmal performance n erms of he orgnal sasfcng objecve. The remander of he paper s srucured as follows. In Secon II we revew he sandard sochasc mul-armed band problem and he assocaed performance bounds. In Secon III we propose he sasfcng objecve and bound performance n erms of hs objecve by defnng a noon of sasfcng regre. In Secon IV we specalze o he case of Gaussan rewards and show ha solvng he sasfcng problem s equvalen o solvng a sandard problem wh Gaussan rewards. In Secon V we revew he UCL algorhm and show how applyng o he problem wh Gaussan rewards acheves opmal performance n erms of he sasfcng objecve. Secon VI shows he resuls of numercal smulaons and Secon VII concludes.

2 II. THE STOCHASTIC MULTI-ARMED BANDIT PROBLEM The sochasc mul-armed band problem s a decsonmakng problem n whch he decson maker sequenally chooses one among a se of N opons, called arms n analogy wh he lever of a slo machne. A sngle-levered slo machne s somemes called a one-armed band, so he case of N opons s ofen called an N-armed band. The decson-makng agen collecs reward r R by choosng arm a each me {1,, T }, where T N s he horzon lengh for he sequenal decson process. The reward from opon {1,, N} s sampled from a saonary probably dsrbuon p and has an unknown mean m R. The decson-maker s objecve s o maxmze some funcon of he sequence of rewards {r }. A. Maxmzaon objecve In he sandard mul-armed band problem, he agen s objecve s o maxmze he expeced cumulave reward [ T T J = E r = m. (1 Equvalenly, by defnng m = max m and R = m m as he expeced regre a me, he objecve (1 can be formulaed as mnmzng he cumulave expeced regre defned by T N R = T m m E [ n T N = E [ n T, ( where n T s he number of mes arm has been chosen up o me T, = m m s he expeced regre due o pckng arm nsead of arm, and he expecaon s over he possble rewards and decsons made by he agen. The nerpreaon of ( s ha subopmal arms should be chosen as rarely as possble. Ths s a non-rval ask snce he mean rewards m are nally unknown o he decson maker, who mus ry all arms o learn abou her rewards whle preferenally pckng arms ha appear more rewardng. The enson beween hese requremens s known as he explore-explo radeoff and s common o many problems n machne learnng and adapve conrol. B. Bound on opmal performance Opmal performance n a band problem wh he maxmzaon objecve (1 corresponds o pckng subopmal arms as rarely as possble, as shown by he las equaly n (. La and Robbns [9 suded he sandard sochasc mul-armed band problem and showed ha any polcy solvng he problem mus pck each subopmal arm a number of mes ha s a leas logarhmc n he me horzon T,.e., E [ ( n T 1 D(p p + o(1 log T, (3 where o(1 0 as T +. The quany D(p p := p (r log p(r p (r dr s he Kullback-Lebler dvergence beween he reward densy p of a subopmal arm and he reward densy p of he opmal arm. The bound on E [ n T mples ha he cumulave expeced regre mus grow a leas logarhmcally n me. The bound (3 s asympoc n me, bu a number of researchers (e.g., [, [5, [13 have consruced algorhms ha acheve cumulave expeced regre ha s bounded by a logarhmc erm unformly n me, somemes wh he same consan as n (3. Cumulave expeced regre ha s unformly bounded n me by a logarhmc erm s ofen called logarhmc regre for shor. In he leraure, algorhms ha acheve logarhmc regre wh a leadng erm ha s whn a consan facor of ha n (3 are consdered o have opmal performance. C. Gaussan rewards In hs paper we focus on he case of Gaussan reward dsrbuons, ha s, he dsrbuon p of rewards assocaed wh arm s Gaussan wh unknown mean m and known varance σs,. In hs case, he Kullback-Lebler dvergence n (3 akes he value ( D(p p = 1 σ s, + σ s, σs, 1 log σ s, σ s,. (4 Ths equaon s more easly nerpreed when he reward varances are unform,.e., σ s, = σ s for each. In hs case, he dvergence becomes so he bound (3 s E [ n T D(p p = σs, ( σ s + o(1 log T. (5 Ths resul can be nerpreed as follows. For a gven value of, a larger varance σ s makes he rewards more varable and herefore s more dffcul o dsngush beween he arms. For a gven value of σ s, a larger value of makes easer o dsngush he opmal arm. III. THE MULTI-ARMED BANDIT PROBLEM WITH SATISFICING OBJECTIVE The sandard mul-armed band problem s defned wh he maxmzng objecve (1. We now propose a new sasfcng objecve for he mul-armed band problem and fnd bounds on opmal performance n erms of hs new objecve. Consder an N-armed band problem. As before, he reward assocaed wh each arm s drawn from a saonary probably dsrbuon p, whose mean m s unknown o he decson maker. A me {1,..., T }, he decson maker selecs arm and receves a sochasc reward r R. The decson maker has a ceran sasfacon level M R, and s sasfed a me only f he reward r s a leas M. Le s be he random varable denong he decson maker s sasfacon a me : { 0, r < M s = 1, r M.

3 Then s s a Bernoull random varable wh success probably π, where π = Pr [s = 1 = = Pr [r M = (6 s he probably of sasfacon upon pckng arm. We propose a sasfcng objecve n erms he number of mes he sasfacon level s me. Defnon 1 (Sasfcng objecve. The sasfcng objecve s o maxmze he funcon [ T T E s = π. (7 The sasfcng objecve dffers from he maxmzaon objecve (1 n several mporan ways. Frs, exhbs hresholdng, ha s, s ndfferen among rewards r above he hreshold value M. Second, exhbs rsk averson, ha s, prefers smaller, conssen rewards (ha wll ofen be above he hreshold o larger, more varable ones (ha may ofen be below. Rsk averson s a characersc ofen suded n economcs and psychology [1, and s ofen ncorporaed n models of human decson makng. Snce he sasfcng objecve consss of maxmzng he number of mes he agen s sasfed, can be rewren as follows. Le π = max π and defne = π π as he expeced sasfcng regre of selecng an arm. We can rewre (7 n erms of expeced sasfcng regre as [ T N J S = E = E [ n T, (8 where n T s he number of mes arm has been chosen up o me T. Ths s a sandard mul-armed band problem wh Bernoull rewards. Therefore he La-Robbns bound (3 holds, yeldng a logarhmc lower bound on E [ n T and cumulave expeced sasfcng regre: Corollary 1 (Sasfcng regre bound. Any polcy solvng he mul-armed band problem wh he sasfcng objecve (8 obeys E [ ( n T 1 D(π π + o(1 log T, (9 for subopmal ( arms ( where D(π π = π π log 1 π π + (1 π log 1 π s he Kullback-Lebler dvergence beween he wo Bernoull dsrbuons wh success probables π and π. Proof: Apply he La-Robbns bound (3 o he sandard mul-armed band problem wh Bernoull rewards. We refer o cumulave expeced sasfcng regre ha s unformly bounded above n me by a logarhmc erm as logarhmc sasfcng regre. An algorhm ha acheves logarhmc sasfcng regre acheves opmal sasfcng performance,.e., opmal performance n erms of he sasfcng objecve (7. The mplcaon of wrng he sasfcng objecve as he mnmzng of cumulave regre s ha f one can use he rewards r o esmae he sasfacon probably π, one can use algorhms desgned o solve he mul-armed band problem wh a maxmzng objecve o solve he sasfcng problem. In he nex secons we sudy he Gaussan mularmed band problem wh a sasfcng objecve and show how o lnk rewards and probables n hs case. IV. SATISFICING WITH GAUSSIAN REWARDS In hs secon we sudy a Gaussan mul-armed band problem wh he sasfcng objecve (8. By Gaussan mularmed band problem, we mean ha he reward r due o selecng arm s r N (m, σs,, where σs, s he known varance of arm. Defne he quany x = m M σ s, (10 for each arm. The followng lemma saes ha he Gaussan mul-armed band problem wh a sasfcng objecve s equvalen o a sandard Gaussan mul-armed band problem wh ransformed reward dsrbuons. Lemma (Equvalence for Gaussan rewards. The Gaussan mul-armed band problem wh sasfcng objecve s equvalen o a sandard Gaussan mul-armed band problem wh rewards r N (x, 1 n he sense ha he orderng of he arms n erms of x s dencal o he orderng n erms of π. In parcular, he arm wh maxmal x s he arm wh maxmal π Proof: Wh Gaussan rewards, he probably (6 of sasfacon from choosng arm s π = Pr [m + σ s, z M ( m M = Φ = Φ(x, σ s, where z N (0, 1 s a sandard normal random varable and Φ(z s s cumulave dsrbuon funcon. Le = arg max π. The key nsgh s ha Φ( s a monooncally ncreasng funcon, whch mples ha he orderng of arms n erms of π s dencal o he orderng n erms of x. In parcular, arm s he arm wh maxmal x. Therefore, he goal of an agen playng he sasfcng band problem s o fnd he arm ha maxmzes x. Ths s agan a Gaussan band problem: consder he ransformed reward r = r M σ s,, whch s a Gaussan random varable r N (x, 1. The quany x plays he role of he mean reward m from he orgnal maxmzng problem and he ransformed rewards have unform varance σ s = 1. Solvng hs problem wh a maxmzng objecve s equvalen o solvng he orgnal problem wh he sasfcng objecve. Remark 3 (Locaon-scale famles. The above analyss s easly generalzed o reward dsrbuons belongng o locaon-scale famles. A locaon-scale famly s a se of

4 probably dsrbuons closed under affne ransformaons,.e., f he random varable X s n he famly, so s he varable Y = a + bx, where a, b R. Any random varable X n such a famly wh mean µ and sandard devaon σ can be wren as X = µ + σz, where Z s a zero-mean, un-varance member of he famly. Examples nclude he unform or Suden s -dsrbuon. V. THE UCL ALGORITHM FOR GAUSSIAN BANDIT PROBLEMS In hs secon we revew he UCL algorhm, a Bayesan algorhm ha we developed and analyzed n [13 o solve he sandard Gaussan band problem. We hen show ha he UCL algorhm can be appled o he Gaussan sasfcng problem of Secon IV, achevng opmal performance. The algorhm manans a belef abou he mean rewards m by sarng wh a pror and updang usng Bayesan nference as new rewards are receved. A each me he algorhm chooses arm usng a heursc whch s a smple funcon of he curren belef sae. For unnformave prors, he UCL algorhm acheves logarhmc regre,.e., opmal performance. Unnformave prors correspond o havng no nformaon abou he mean rewards. A major aspec of he UCL algorhm s s ably o ncorporae nformaon abou he mean rewards hrough he use of a so-called nformave pror. In [13, we show ha an appropraely chosen pror can sgnfcanly ncrease he performance of he UCL algorhm. Several dfferen UCL algorhms are developed n [13; here we cover only he deermnsc UCL algorhm, whch we refer o as he UCL algorhm for brevy. A. Pror The pror dsrbuon capures he agen s knowledge abou he vecor of mean rewards m before begnnng he ask. We assume ha he pror dsrbuon s mulvarae Gaussan wh mean µ 0 R N and covarance Σ 0 R N N : m N (µ 0, Σ 0. (11 The h elemen of µ 0, denoed by µ 0, represens he agen s mean belef of he reward m assocaed wh arm. The (, elemen of Σ 0, denoed by (, σ 0 represens he agen s uncerany assocaed wh ha belef. Off-dagonal elemens of Σ 0, e.g., σj 0, represen he agen s perceved relaonshp beween m and m j : f σj 0 s posve, hgh values of m are correlaed wh hgh values of m j, whle f s negave, hgh values of m correlae wh low values of m j. Any posvedefne marx can be used as Σ 0, bu several specfc ones are of neres. An unnformave pror corresponds o a complee lack of cerany,.e., ( σ 0 +, so one ses each elemen σj 0 equal o +. B. Inference updae A each me he agen pcks an arm and receves a reward r ha s Gaussan dsrbued: r N (m, σ s,. Bayesan nference provdes an opmal soluon o he problem of updang he belef sae (µ, Σ o ncorporae hs new nformaon. Gven he Gaussan pror (11, he Bayesan updae equaons are lnear [8: q = r φ σs, + Λ 1 µ 1 Λ = φ φ T σs, µ = Σ q. C. Decson heursc + Λ 1, Σ = Λ 1 (1 A each me he UCL algorhm compues a value Q for each arm. The UCL algorhm pcks he arm ha maxmzes Q. Tha s, pcks The heursc value Q s = arg max Q. (13 Q = µ + σ Φ 1 (1 α, (14 where µ = (µ, (σ = (Σ, α = 1/K, and K > 0 s a unable parameer. The heursc Q s a Bayesan upper lm for he value of m based on he nformaon avalable a me. I represens an opmsc assessmen of he value of m. The decson made can be hough of as he mos opmsc one conssen wh he curren nformaon. D. Performance In [13, we sudy he case of homogeneous samplng nose (.e., σ s, = σ s for each and show ha he UCL algorhm acheves cumulave expeced regre unformly n me. In parcular, we prove ha he followng heorem holds for any β 1.0. Theorem 4 (Regre of he deermnsc UCL algorhm [13. The followng saemens hold for he Gaussan mularmed band problem and he deermnsc UCL algorhm wh uncorrelaed unnformave pror and K = : 1 he expeced number of mes a subopmal arm s chosen unl me T sasfes E [ n T ( 8β σ s + log T + 4β σs (1 log log log T ; he cumulave expeced regre unl me T sasfes ( T N (8β σs R + log T + 4β σs (1 log log log T The mplcaon of hs heorem can be seen by comparng saemen 1 wh he La-Robbns bound (5: shows ha he UCL algorhm acheves logarhmc regre unformly n me wh a consan ha dffers from he opmal asympoc one by a consan facor of 4β, and herefore s consdered o have opmal performance.

5 E. Applcaon o sasfcng objecve In Secon IV, we showed ha solvng he Gaussan mul-armed band problem wh a sasfcng objecve s equvalen o a ransformed sandard Gaussan mul-armed band problem wh maxmzng objecve. Therefore, we can apply he UCL algorhm o he sasfcng problem. A pror belef m N (µ 0, Σ 0 s ransformed no pror belefs on x by x N ( µ 0, Σ 0, where ( µ 0 = ((µ 0 M/σ s, and ( Σ 0 j = (Σ 0 j /(σ s, σ s,j. Defne x = max x and = x x. We refer o he UCL algorhm usng he ransformed reward r and pror as he sasfcng UCL algorhm. The sasfcng UCL algorhm acheves logarhmc sasfcng regre, as formalzed n he followng heorem. Theorem 5 (Regre of he sasfcng UCL algorhm. The followng saemens hold for he Gaussan mul-armed band problem wh a sasfcng objecve and he sasfcng UCL algorhm wh uncorrelaed unnformave pror and K = : 1 he expeced number of mes a subopmal arm s chosen unl me T sasfes E [ n T (8β + log T + 4β (1 log log log T ; he cumulave expeced sasfcng regre unl me T sasfes T N R ( (8β + log T (15 + 4β (1 log log log T Proof: Apply Theorem 4 o he Gaussan mul-armed band problem wh mean rewards x and reward dsrbuons r N (x, 1 defned n Lemma. The sasfcng regre s upper bounded by a logarhmc funcon of T. Therefore, he sasfcng UCL algorhm acheves opmal sasfcng regre up o a consan facor. VI. NUMERICAL EXAMPLE In hs secon, we presen he resuls of wo numercal smulaons of he sasfcng UCL algorhm solvng a mul-armed band problem wh Gaussan rewards and he sasfcng objecve. The frs smulaon demonsraes he performance guaranees and allows us o compare he opmal regre bound (9 and he bound (15 obeyed by he sasfcng UCL algorhm. The second smulaon demonsraes he rsk-averse naure of he sasfcng objecve. For he smulaons presened n Fgure 1, we se N = 4. The sasfacon level M was se equal o, he mean rewards m were equal o [1 3 4 and he sandard devaons equal o [ , so x = [ and = 3 was he opmal arm. The algorhm used an unnformave pror. These values were chosen such ha he arm wh maxmal mean reward was no he opmal arm, so sasfcng nduces dfferen behavor han maxmzng. Fgure 1 plos he mean cumulave sasfcng regre ncurred by he sasfcng UCL algorhm over 100 smulaons along wh he wo regre bounds (9 and (15. The mean regre obeys he performance bound (15 from Theorem 5 and s acually below he asympoc lower bound (9 a nal mes. Ths apparen volaon of he bound s due o he fac ha a nal mes he sysem s no ye n he asympoc regme where he bound apples. For he smulaons presened n Fgure, we se N =. The mean rewards m were equal o [ and he sandard devaons equal o [10 1, so x = [ Ths mean = 1 was he opmal arm for he maxmzng objecve whle = was he opmal arm for he sasfcng objecve. The algorhm used an unnformave pror. The problem was smulaed 100 mes wh each objecve. Fgure demonsraes he rsk averson nheren n he sasfcng objecve by comparng he resuls of he same problem solved wh he sasfcng and he maxmzng objecves. The sasfacon level M was se equal o 1. We consdered cumulave surplus (rewards n excess of he sasfacon level for boh objecves. Negave values of he surplus represen defcs, whch are o be avoded. Resuls from he maxmzng objecve are presened n black. The sold lne shows mean cumulave surplus and he shaded regon shows he 95% confdence nerval around ha mean. Resuls from he sasfcng objecve are presened n blue. The sold lne shows he mean cumulave surplus, and he dashed lnes show he 95% confdence nerval. The lower lm of he confdence nervals measures wors-case performance. The measure for he sasfcng objecve s conssenly above he one for he maxmzng objecve, so sasfcng resuls n beer wors-case performance. VII. CONCLUSION Sasfcng, he concep of dong well relave o a reference value, s a useful alernave o maxmzng ha can be appled o a varey of decson-makng scenaros. Consderng sasfcng objecves nsead of maxmzng ones can smplfy decson-makng problems and can resul n polces ha are more robus n he sense ha hey are rsk-averse. In hs paper, we consdered he mul-armed band problem usng a sasfcng objecve by proposng a new noon of sasfcng regre. We showed ha here s an equvalence beween mnmzng sasfcng regre and mnmzng he sandard noon of regre. Usng hs equvalence, we derved a logarhmc lower bound on sasfcng regre and, n he case of Gaussan rewards, adaped he UCL algorhm [13 o acheve opmal sasfcng performance. Ths work opens he door o many fuure exensons. The sasfcng objecve wh Gaussan rewards bears a srong resemblance o he CredMercs wo-sae cred rsk model used n quanave fnance [7. Ths could allow he cred

6 Cumulave regre Cumulave surplus Fg. 1. Regre ncurred by he sasfcng UCL algorhm whle solvng a sasfcng Gaussan mul-armed band problem, along wh wo heorecal bounds, ploed agans me on a logarhmc scale. The sold black lne shows mean cumulave expeced regre from 100 smulaons. The dashed lne shows he asympoc bound on regre (9, whch appears as a sragh lne due o he scalng of he axes. The dash-doed lne shows he regre bound (15, whch provdes guaranees on he algorhm s performance. nvesmen porfolo problem suded n fnance o be posed as a mul-armed band problem wh sasfcng objecve. The rsk averse naure of sasfcng objecves such as he one proposed n hs paper wll resul n more robus polces for solvng he mul-armed band problem n cases wh reward varance σs s heerogeneous across arms. Rsk averson and robusness are mporan for engneerng applcaons (where sandard band algorhms are known o have poor rsk-averson characerscs [1 bu also n he feld of opmal foragng heory [4. The mul-armed band framework has been used o sudy foragng [18 usng a maxmzng objecve, bu a sasfcng objecve s more ecologcally plausble. We developed a polcy for he sasfcng problem wh Gaussan rewards, bu developmen of opmal polces for he sasfcng problem wh oher reward dsrbuons remans an open problem. For all sasfcng problems, pckng he approprae sasfacon level s a non-rval problem n s own rgh, analogous o pckng he error raes n he Sequenal Probably Rao Tes [19. ACKNOWLEDGEMENT We hank Vabhav Srvasava and Smon A. Levn for helpful dscussons. REFERENCES [1 J.-Y. Audber, R. Munos, and C. Szepesvár. Exploraon exploaon radeoff usng varance esmaes n mul-armed bands. Theorecal Compu. Sc., 410(19: , 009. [ P. Auer, N. Cesa-Banch, and P. Fscher. Fne-me analyss of he mularmed band problem. Mach. Learnng, 47(:35 56, 00. [3 R. Bordley and M. LCalz. Decson analyss usng arges nsead of uly funcons. Decsons n Economcs and Fnance, 3(1:53 74, 000. [4 Y. Carmel and Y. Ben-Ham. Info-gap robus-sasfcng model of foragng behavor: Do foragers opmze or sasfce? The Amer. Naurals, 166(5: , Fg.. Cumulave surplus earned by he UCL algorhm whle solvng a Gaussan mul-armed band problem, once wh a sasfcng (blue curves and agan wh a maxmzng objecve (black curve and shaded regon. Boh objecves acheve smlar mean performance (sold curves bu usng he sasfcng objecve resuls n beer wors-case performance. The shaded regon (sasfcng and he blue dashed lnes (maxmzng show he 95% confdence nerval around he mean cumulave surplus. The lower lm of he confdence nervals measures wors-case performance. The lower lm for he sasfcng objecve s conssenly above he one for he maxmzng objecve, so sasfcng resuls n beer wors-case performance. [5 A. Garver and O. Cappé. The KL-UCB algorhm for bounded sochasc bands and beyond. In JMLR: Workshop and Conference Proceedngs, volume 19: COLT 011, pages , 011. [6 M. Goodrch, W. Srlng, and R. Fros. A heory of sasfcng decsons and conrol. IEEE Trans. Sys., Man and Cybern. A: Sys. Humans, 8(6: , Nov [7 M. B. Gordy. A comparave anaomy of cred rsk models. J. of Bankng & Fnance, 4(1: , 000. [8 S. M. Kay. Fundamenals of Sascal Sgnal Processng, Volume I : Esmaon Theory. Prence Hall, [9 T. L. La and H. Robbns. Asympocally effcen adapve allocaon rules. Advances n Appl. Mah., 6(1:4, [10 T. M. Moe. The new economcs of organzaon. Amer. J. of Polcal Sc., 8(4: , [11 H. Nakayama and Y. Sawarag. Sasfcng rade-off mehod for mulobjecve programmng. In Ineracve Decson Analyss, pages Sprnger, [1 J. W. Pra. Rsk averson n he small and n he large. Economerca: J. of he Economerc Soc., pages 1 136, [13 P. Reverdy, V. Srvasava, and N. E. Leonard. Modelng human decson-makng n generalzed Gaussan mul-armed bands. Proc. IEEE, 10(4: , 014. [14 H. Robbns. Some aspecs of he sequenal desgn of expermens. Bullen of he Amer. Mah. Soc., 58:57 535, 195. [15 B. Schwarz, A. Ward, J. Monerosso, S. Lyubomrsky, K. Whe, and D. R. Lehman. Maxmzng versus sasfcng: happness s a maer of choce. J. Personaly and Socal Psychology, 83(5:1178, 00. [16 H. A. Smon. A behavoral model of raonal choce. The Quarerly J. of Econ., 69(1:99 118, [17 H. A. Smon. Raonal choce and he srucure of he envronmen. Psychologcal Revew, 63(:19, [18 V. Srvasava, P. Reverdy, and N. E. Leonard. On opmal foragng and mul-armed bands. In Proc. of he 51s Annu. Alleron Conf. on Commun., Conrol, and Compung, pages , 013. [19 A. Wald. Sequenal ess of sascal hypoheses. Annals of Mahemacal Sascs, 16(: , [0 D. Ward. The role of sasfcng n foragng heory. Okos, pages , 199. [1 S. G. Wner. The sasfcng prncple n capably learnng. Sraegc Managemen Journal, 1(10-11: , 000. [ B. Yn e al. Fndng opmal soluon for sasfcng non-funconal requremens va 0-1 programmng. In Proc. 013 IEEE 37h Annu. Compuer Sofware and Applcaons Conf., pages , 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

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

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

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

On Optimal Foraging and Multi-armed Bandits

On Optimal Foraging and Multi-armed Bandits 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

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

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

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

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

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

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

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

( ) [ ] 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

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

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

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

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

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

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

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

More information

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

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

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

Volatility Interpolation

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

More information

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

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

( ) () 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

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

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

More information

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

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

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

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

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

Advanced time-series analysis (University of Lund, Economic History Department)

Advanced time-series analysis (University of Lund, Economic History Department) Advanced me-seres analss (Unvers of Lund, Economc Hsor Dearmen) 3 Jan-3 Februar and 6-3 March Lecure 4 Economerc echnues for saonar seres : Unvarae sochasc models wh Box- Jenns mehodolog, smle forecasng

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

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

Lecture 6: Learning for Control (Generalised Linear Regression)

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

More information

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

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

Lecture VI Regression

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

More information

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

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

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

More information

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

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

More information

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

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

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

More information

Bayes rule for a classification problem INF Discriminant functions for the normal density. Euclidean distance. Mahalanobis distance

Bayes rule for a classification problem INF Discriminant functions for the normal density. Euclidean distance. Mahalanobis distance INF 43 3.. Repeon Anne Solberg (anne@f.uo.no Bayes rule for a classfcaon problem Suppose we have J, =,...J classes. s he class label for a pxel, and x s he observed feaure vecor. We can use Bayes rule

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

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

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

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

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

More information

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

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

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

More information

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

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

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

Math 128b Project. Jude Yuen

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

More information

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

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

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

Standard Error of Technical Cost Incorporating Parameter Uncertainty

Standard Error of Technical Cost Incorporating Parameter Uncertainty Sandard rror of echncal Cos Incorporang Parameer Uncerany Chrsopher Moron Insurance Ausrala Group Presened o he Acuares Insue General Insurance Semnar 3 ovember 0 Sydney hs paper has been prepared for

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

Advanced Machine Learning & Perception

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

More information

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

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

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

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

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

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

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

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

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

More information

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

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

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

Analysis And Evaluation of Econometric Time Series Models: Dynamic Transfer Function Approach

Analysis And Evaluation of Econometric Time Series Models: Dynamic Transfer Function Approach 1 Appeared n Proceedng of he 62 h Annual Sesson of he SLAAS (2006) pp 96. Analyss And Evaluaon of Economerc Tme Seres Models: Dynamc Transfer Funcon Approach T.M.J.A.COORAY Deparmen of Mahemacs Unversy

More information

P R = P 0. The system is shown on the next figure:

P R = P 0. The system is shown on the next figure: TPG460 Reservor Smulaon 08 page of INTRODUCTION TO RESERVOIR SIMULATION Analycal and numercal soluons of smple one-dmensonal, one-phase flow equaons As an nroducon o reservor smulaon, we wll revew he smples

More information

[Link to MIT-Lab 6P.1 goes here.] After completing the lab, fill in the following blanks: Numerical. Simulation s Calculations

[Link to MIT-Lab 6P.1 goes here.] After completing the lab, fill in the following blanks: Numerical. Simulation s Calculations Chaper 6: Ordnary Leas Squares Esmaon Procedure he Properes Chaper 6 Oulne Cln s Assgnmen: Assess he Effec of Sudyng on Quz Scores Revew o Regresson Model o Ordnary Leas Squares () Esmaon Procedure o he

More information

NPTEL Project. Econometric Modelling. Module23: Granger Causality Test. Lecture35: Granger Causality Test. Vinod Gupta School of Management

NPTEL Project. Econometric Modelling. Module23: Granger Causality Test. Lecture35: Granger Causality Test. Vinod Gupta School of Management P age NPTEL Proec Economerc Modellng Vnod Gua School of Managemen Module23: Granger Causaly Tes Lecure35: Granger Causaly Tes Rudra P. Pradhan Vnod Gua School of Managemen Indan Insue of Technology Kharagur,

More information

Tools for Analysis of Accelerated Life and Degradation Test Data

Tools for Analysis of Accelerated Life and Degradation Test Data Acceleraed Sress Tesng and Relably Tools for Analyss of Acceleraed Lfe and Degradaon Tes Daa Presened by: Reuel Smh Unversy of Maryland College Park smhrc@umd.edu Sepember-5-6 Sepember 28-30 206, Pensacola

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

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

[ ] 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

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

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

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

Dynamic Team Decision Theory

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

More information

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

Lecture 18: The Laplace Transform (See Sections and 14.7 in Boas)

Lecture 18: The Laplace Transform (See Sections and 14.7 in Boas) Lecure 8: The Lalace Transform (See Secons 88- and 47 n Boas) Recall ha our bg-cure goal s he analyss of he dfferenal equaon, ax bx cx F, where we emloy varous exansons for he drvng funcon F deendng on

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

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

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

More information

Lecture Notes 4. Univariate Forecasting and the Time Series Properties of Dynamic Economic Models

Lecture Notes 4. Univariate Forecasting and the Time Series Properties of Dynamic Economic Models Tme Seres Seven N. Durlauf Unversy of Wsconsn Lecure Noes 4. Unvarae Forecasng and he Tme Seres Properes of Dynamc Economc Models Ths se of noes presens does hree hngs. Frs, formulas are developed o descrbe

More information

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

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

More information

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

THE PREDICTION OF COMPETITIVE ENVIRONMENT IN BUSINESS

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

More information

Time Scale Evaluation of Economic Forecasts

Time Scale Evaluation of Economic Forecasts CENTRAL BANK OF CYPRUS EUROSYSTEM WORKING PAPER SERIES Tme Scale Evaluaon of Economc Forecass Anons Mchs February 2014 Worng Paper 2014-01 Cenral Ban of Cyprus Worng Papers presen wor n progress by cenral

More information

Mechanics Physics 151

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

More information

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

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

RELATIONSHIP BETWEEN VOLATILITY AND TRADING VOLUME: THE CASE OF HSI STOCK RETURNS DATA

RELATIONSHIP BETWEEN VOLATILITY AND TRADING VOLUME: THE CASE OF HSI STOCK RETURNS DATA RELATIONSHIP BETWEEN VOLATILITY AND TRADING VOLUME: THE CASE OF HSI STOCK RETURNS DATA Mchaela Chocholaá Unversy of Economcs Braslava, Slovaka Inroducon (1) one of he characersc feaures of sock reurns

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

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