Global Search in Combinatorial Optimization using Reinforcement Learning Algorithms

Size: px
Start display at page:

Download "Global Search in Combinatorial Optimization using Reinforcement Learning Algorithms"

Transcription

1 Global Search n Combnatoral Optmzaton usng Renforcement Learnng Algorthms Vctor V. Magkkh and Wllam F. Punch III Genetc Algorthms Research and Applcaton Group (GARAGe) Mchgan State Unversty 2325 Engneerng Buldng East Lansng, MI Phone: (517) E-mal: {magkkh,punch}@cse.msu.edu Abstract- Ths paper presents two approaches that address the problems of the local character of the search and mprecse state representaton of renforcement learnng (RL) algorthms for solvng combnatoral optmzaton problems. The frst, Bayesan, approach ams to capture soluton parameter nterdependences. The second approach combnes local nformaton as encoded by typcal RL schemes and global nformaton as contaned n a populaton of search agents. The effectveness of these approaches s demonstrated on the Quadratc Assgnment Problem (QAP). Compettve results wth the RL-agent approach suggest that t can be used as a bass for global optmzaton technques. 1 Introducton The success of a generate-and-test algorthm for a partcular problem s determned by factors such as: ts ablty to use past experence to form feasble soluton ts explotaton/exploraton strategy, ts utlzaton of problemspecfc nformaton etc. In creatng a feasble soluton, the algorthm has to make a number of decson e.g. whch value should be assgned to a partcular free parameter. The qualty of the soluton generated s often the only type of feedback avalable after a sequence of decsons s made. Snce we expect the algorthm to make decsons whch result n better solutons over tme, the problem of ntellgent soluton generaton can be approached wth renforcement learnng (RL). The problems wth delayed renforcement that RL approaches face are well modeled by Markov Decson Processes (MDPs). MDPs are defned by: a set of Markov state the actons avalable n those state the transton probabltes and the rewards assocated wth each stateacton par. Model based RL-algorthms are explctly lookng for MDP soluton, an optmal polcy, whch s a mappng from MDP states to actons whch maxmzes the expected average reward receved by followng a path through MDP states. An acton-value functon for a polcy s defned as a mappng from each state-acton par to the expected average reward obtaned by choosng an acton n that state accordng to the gven polcy and followng that polcy thereafter. The state-value functon for a polcy specfes the desrablty of a state and s defned as the expected average reward obtaned by followng that polcy from a gven state. Snce probabltes are not always known, typcal RL algorthm e.g. SARSA or Q-learnng, are model free. Iteratve updates used by these algorthms do not use transton probabltes and are proven to converge to optmal value functon. A greedy polcy that chooses an acton accordng to the maxmum optmal acton-value s known to be globally optmal based on the expected average reward crteron. Readers nterested n a more detaled treatment of RL should read references such as Sutton and Barto (1997). For an example of an optmzaton problem formulated n RL term consder an RL approach to the Travelng Salesman Problem (TSP). The states are the ctes. The actons are the choces of the next cty to vst, and the acton-values ndcate the desrablty of the cty to vst next. Global reward s the nverse of the tour length. Immedate rewards can be defned as nverse of the dstance between a par of ctes. The dea of usng RL n optmzaton problem solvng s almost as old as RL tself. It was frst studed n the n- armed bandt by Bellman (1956) and later appled to more dffcult optmzaton problems by varous researchers. For example, Dorgo (1992,1996) has developed an optmzaton technque known as Ant Systems(AS). The key dea behnd AS s a heurstc approxmaton to actonvalue whch he terms pheromone. Even though AS were derved by smulatng the behavor of a populaton of ant they have much n common wth other RL algorthms. Another applcaton of RL to optmzaton s that of Crtes and Barto (1996) where they appled Q-learnng to elevator schedulng. Ths paper s partcularly relevant to our research snce t explores the possbltes of mult-agent RL algorthms n optmzaton. Each agent n a team of RL algorthms controls a partcular elevator car cooperatvely solvng the entre problem. Among other relevant publcaton Gambardella and Dorgo (1995) who

2 descrbed the applcaton of Q-learnng to TSP and asymmetrc TSP. Zhang and Detterch (1996) use TD(λ) to solve a Job-Shop Schedulng problem. Sngh and Bertsekas (1996) used RL for the channel allocaton problem. In order to dscuss advantages and dsadvantages of RL n optmzaton, let us frst contrast them aganst another well-known optmzaton technque, genetc algorthms (GA). Whle RL supports value-functon, whch reflects the desrablty of free parameter assgnment GA approaches explctly vew only the overall ftness of a soluton. In constructng a new soluton, GAs are not guded by any synthetc ftness values assocated wth any smaller part of soluton. Rather, GAs are guded by schema theory whch states that the more favorable a partcular choce of values for a subset of soluton parameters the more frequently such a schema appears as a part of solutons n the populaton. These buldng blocks thus represent the preferred values of soluton parameters and ther combnatons. The ablty to both explore and explot schemata n the search space s the key to GA success as frst ponted out by Holland (1975). Thu each schema n a GA has an mplct probablty of appearng n generated soluton where the better a schema the hgher the probablty of t occurrng n a soluton. Such a representaton s smlar to tossng a con and storng all outcomes nstead of the number of trals and the number of heads. Ths rases the queston: s a populaton of solutons an accurate and computatonally effectve way of representng the preferences of free parameter choces as compared to some form of suffcent statstcs? Snce the number of schemata grows exponentally wth the sze of the problem, mantanng values assocated wth each possble combnaton of parameters becomes prohbtve. On the other hand, GAs do not drectly learn from bad experence. Moreover, fnte populatons can drop some alleles from the populaton and there s only a slght chance that they may be rentroduced va mutaton, and they may not survve to be used. Use of RL technques n optmzaton problems has good and bad aspects. On the postve sde, they are proven to converge to optmum gven the rght crcumstances and are applcable to problems wth a large number of states. They can also be used n conjuncton wth functonapproxmaton technques to add generalzaton and reduce space requrements. Boyan and Moore (1998) report good results on a number of dscrete and contnuous optmzaton problems usng ths approach. Drect estmaton of desrablty of assgnments by value functons has a potental to be both more precse and computatonally cheaper than other approaches. Ths possblty s one of the major motvatons for conductng research on applcablty of RL algorthms to optmzaton. There are also dsadvantages. The frst s the local rather than global character of search n the RL schemes proposed so far. The algorthm has to explore the space by choosng probablstcally from among all acton not just the acton wth the hghest acton-value estmate. In combnatoral optmzaton, even one ncorrect exploratory step can serously damage the qualty of the resultant soluton. Therefore, to generate a good soluton, the most preferable acton has to be selected most of the tme, whch strongly shfts the balance from exploraton to explotaton and leads to local rather than global search. Another problem n RL s the coarse representaton of the state. For nstance, n solvng a TSP by AS as descrbed n Dorgo et al. (1996) or Ant-Q n Gambardella and Dorgo (1995), the state s the current cty, and the acton s whch cty to vst next. Clearly a full representaton of the state would contan both the current cty and the tour of ctes already vsted. Snce ths hstory obvously nfluences further assgnment, ther smple defnton loses the Markov property, and a suboptmal sequence of cte a buldng block, s not captured. Consequently, the algorthm wll not be able to handle parameter nterdependence suffcently well. As mentoned earler, the number of states n an RL approach cannot be so large as to keep an estmate for every possble sequence because the number of states grows exponentally. Nevertheles ths problem may be addressed by the use of functonapproxmaton and other means as wll be dscussed further. 2 Capturng Parameter Interdependences Usng a Bayesan Approach The coarse state representaton does not take nto account nteracton of avalable assgnments wth those already made. To nclude these nteractons we can use a Bayesan approach. To construct a feasble soluton for a combnatoral optmzaton problem, a number of free parameters should be nstantated. Let x ψ denote the fact that some free parameter x s assgned value ψ. For example, n the Quadratc Assgnment Problem descrbed n secton 4, the free parameters are locatons and the values to be assgned to those parameters are ordnal numbers of facltes. Let P ( y χ x ψ ) denote condtonal probablty of assgnng a free parameter y the value χ gven that x was already assgned ψ. Ths condtonal probablty ndcates an assgnment already made nfluences the probablty of the assgnment under consderaton. We can fnd P ( y χ x ψ ) usng Bayes rule: x ψ y χ ) P ( y χ x ψ ) = (2.1) x ψ ) where P ( y χ) s the pror probablty of assgnng y to χ and P ( x ψ s the lkelhood of assgnment

3 y χ wth respect to assgnment x ψ. If we set P ( x ψ ) to one, (2.1) can be smplfed to P ( y χ x ψ ) = x ψ y χ). Now suppose that a sequence of k assgnments S ( x1 ψ 1, x2 ψ 2,..., x k ψ k ) was made. The posteror probablty of assgnment y χ gven that assgnments n S are made s P ( y χ S) = S y χ). By computng ths posteror probablty for all canddate values of χ for free parameter y n consderaton, one can fnd how partcular χ fts wth assgnments already made. Note that the number of condtonal probabltes P S x ψ, x ψ,..., x ψ y ) ( k k χ grows exponentally wth the sze of the problem. We can use the nave Bayesan approach to address ths problem. Assumng ndependence of the assgnments x ψ we get: k y χ S) = y χ) x ψ (2.2) = 1 The probablty of assgnment y χ gven that k assgnments n S were already made s the product of ts pror probablty P ( y χ), and condtonal probabltes x ψ for all pror assgnments. If n s the total number of decsons to construct a feasble soluton, then there are O(n 4 ) condtonal probabltes x ψ. Thu even the nave Bayesan scheme leads to a hgh, O(n 4 ), space complexty, whch s acceptable only for moderately szed problems. Snce pror and condtonal probabltes are not known, ther estmates P ˆ( y χ) and Pˆ ( x ψ should be determned n the course of the search. Those can be found as the frequences of assgnments co-occurrences. However, mantanng both probablty estmates and valuefuncton estmate nvolves sgnfcant ncrease of space and computatonal requrement. To avod th we may keep only value functon estmates usng dependence of actons probabltes on acton-values and polcy beng used. Under any reasonable polcy, the actons wth hgher acton-values have more chance of beng selected. From RL pont of vew, a state corresponds to a free parameter and an acton corresponds to a choce of value for that parameter. For proportonal polcy, the probablty estmate, P ˆ( s, a ), of acton a n state s proportonal to ts acton-value, Q ˆ( s, a ) : Pˆ( a) = Qˆ( a) Qˆ( a' ) a' A Pˆ( a) Qˆ( a) (2.3) where A s the set of all actons avalable n s. Assumng (2.3) for all possble polce we can use Bayesan scheme not on the probablty estmate but on normalzed actonvalues. For example, usng a Monte Carlo update rule, we can fnd the acton-values of assgnments as: ~ P ( y χ) = P ~ ( y χ) + α[ r P ~ ( y χ)] (2.4) P ~ ( x ~ ψ ) ( ) [ y χ = P x ψ y χ + + β r x ψ ] (2.5) where ~ P ( y χ) Qˆ( y, χ) Pˆ( y, χ) Pˆ( y χ ) to conform wth our notaton for assgnments and show the relatonshp wth probabltes. P ~ ( x ψ s the expected average reward for takng acton ψ n state gven that acton χ was taken n state y. The reward r can be based on the comparson wth the average ftnes avft, of the last M solutons generated: r = ( avft Ftness) / avft (2.6) The hgh space complexty of ths approach s a problem. Ths approach s therefore only possble for problem nstances of moderate sze unless used along wth functonapproxmaton. Also, snce only n 2 of the total of O(n 4 ) ~ entres x ψ get updated per teraton, ths method wll converge slowly. However, snce ths approach attempts to solve the problem of parameter nterdependence drectly, t has sgnfcant theoretcal mportance for the sake of comparson wth ndrect approache one of whch wll be ntroduced further. 3 The Approach Based on a Populaton of RL Search Agents Snce drect capturng of parameter nterdependences by keepng addtonal estmates s expensve, we can thnk about ndrect approaches. One such approach s as follows: we contnue to use a coarse representaton of the state but stop lookng for general preference values whch would be vald n any part of the search space. Snce coarse representaton collapses many true states of the system nto one makng them ndstngushable, the acton-values assocated wth coarse state -acton pars can only be vald for a local part of the search space. We wll call ths the prncple of localty of acton-values. However, actonvalues from dfferent parts of the search space can be more broadly applcable. Therefore, ths approach mantans a populaton of not only soluton whch are the best results of the search conducted by the RL algorthm stuated n some area of the search space, but also ther acton-values. Ths couplng of a locally-best soluton, the acton-values and an RL algorthm s defned as an agent, an expert n ts local area of the search space. As soon as we have local nformaton from dfferent parts of the search space, we x

4 need a way to combne the results of best yet search n one area wth another. Snce each agent n the populaton s addressng the same optmzaton problem, we expect that at least some other agent s acton-values are useful n areas other than the local space n whch there were formed. Ths assumpton of homogenety allows us to combne results from multple agents. Consder one such approach: a new soluton s formed by copyng a part of the locally-best soluton found by one agent, whle the remanng assgnments are made usng acton-values borrowed from another agent. How would ths compare to recombnng two solutons usng GA crossover? In GA crossover we have two knds of nformaton, the two nstances and perhaps some problemspecfc nformaton. For example, Grefenstette (1985) crossover for TSP has to make 40% of ts assgnments at random to avod conflcts wth prevous assgnments. Wth acton value we can drect those assgnments rather than make them randomly. Ths ncreases the chances of fndng a good sequence. Thu the operaton descrbed looks lke a knd of crossover, usng two nstances to generate one chld, based on ndrect transfer of nformaton though acton-values. We may also thnk of t as combnng both partal results and preferences resultng from search conducted by other agents. Possble varaton of ths theme s to generate a partal soluton wth one agent and use another agent to generate the remander. Approaches usng both a central soluton and acton-values are also possble. Ths synthetc approach would allow combnng the advantages of both RL and GA. In addton to capturng nterdependence a populaton of RL search agents provdes oportuntes for more global search. As was noted n the ntroducton, the local character of the search comes n part from constructng the entre soluton from scratch. Our approach uses an RL algorthm to generate not the whole soluton, but only a part of t. The other part s replcated from the best soluton found so far by ths or another RL algorthm. At frst glance ths mght seem to make the approach even more locally orented. Ths s the case only f the replcated part s dscovered by some other agent, whch followed a smlar thread of search. To enforce ndependent threads of search as conducted by each agent n the populaton, we can choose the followng replacement polcy: the chld competes wth the parent whch was the source of replcated materal, and the better soluton (parent or chld) s placed nto the next generaton. In ths case, two agents are smlar (same preference etc.) only f they dscovered the same soluton ndependently based on ther own acton-values. Another way to make the search more global s to allow the RL approach to wander more (follow less strngently ts preferences). To avod ntroducng poor solutons nto the populaton, each soluton can be passed through a problem-specfc local optmzer to see f ths exploraton found a useful area of the search space. These two approaches are complementary because ndependent threads reduce crowdng whch can cause prelmnary convergence to a local optmum. In ts turn, local optmzaton allows broader search by allowng parameters controllng exploraton n the RL algorthm to be set less tghtly. Snce an nstance n the populaton s not only a soluton, but also a matrx of acton-value t s costly to copy. Ths s one of the reasons that compettve replacement s used n the algorthm. We assume here that f the chld s better than the parent whch served as the source of replcated part, then the chld nherts all the preference values of that parent. Dependng on the results of competton, the update of the preference values s made ether n both parents or n the chld and the parent and there s no need to copy them. Intalze populaton and parameters; Repeat Select two agents A 1 and A 2 from the populaton usng e.g. proportonal selecton based on the ftness of central soluton; For each free parameter wth probablty λ do Copy the value of free parameter from A 1 to offsprng O; End For each unassgned free parameter n O do In problem specfc order: Select a value to be assgned to ths free parameter from the set of possble value accordng to some polcy based on the acton-values of A 2 and assgn t to that free parameter; End Pass O through local optmzer (optonal step); Evaluate O; f(o) denotes ftness of O; Compute reward r; If f(o) s better then the ftness of central soluton of A 1 then Copy O to central soluton of A 1; End Update acton-values of A 1 and A 2 usng reward r; Untl termnaton condton; Output best soluton n populaton Fgure 1: Hgh level pseudocode for RL-agent approach. The hgh-level pseudocode of the approach s shown n Fgure 1. There s a populaton of RL agents where each s comprsed of a locally best soluton, a matrx of actonvalues and the parameters for the RL algorthm. To produce a new agent, two solutons are selected from a populaton usng proportonal or another type of selecton. The new soluton s formed usng the soluton of one parent and the acton-values of the other. After calculatng the ftness of the new soluton, the chld competes wth the

5 parents for ncluson n the populaton. Then the valuefunctons are updated that completes the generaton cycle. The reward could be based on the dfference of the ftness of the new soluton and the average ftness of the parents or some other baselne. Dependng on the problem beng solved and the partcular RL algorthm used, local rewards could also be employed. 4 Applcaton to the QAP The Quadratc Assgnment Problem s a NP-hard problem of fndng a permutaton ϕ mnmzng: Z = n n C ϕ + A B (4.1) ( ) j ϕ( ) ϕ ( j) = 1 = 1 j= 1 where n s the number of facltes/locaton C j cost of locatng faclty at locaton j, A j s cost of transferrng a materal unt from locaton to locaton j, B j s the flow of materal from faclty to faclty j. The permutaton ϕ ndcates the assgnment of a faclty to a locaton. The double summaton of the products term makes the QAP hghly non-lnear. The preference values can estmate the goodness of assgnng a specfc locaton to some faclty. The result of assgnng faclty to locaton s hghly dependent on how other facltes are assgned. Ths property makes ths problem to be very nterestng subject for testng the presented approaches on. 4.1 Populaton-Based Approach In accordance wth the approach, the new feasble soluton s formed n part by replcatng the fragments of the best soluton dscovered by one of the agents and fllng n the remanng part usng the value functon of another agent. To construct a new feasble soluton, unoccuped locatons were selected n random order and assgned facltes usng ε -greedy proportonal polcy. Ths polcy wth the probablty ε chooses the faclty from the set of not-yetassgned facltes havng maxmum acton-value, or wth probablty 1-ε, chooses one of the remanng optons wth probablty proportonal to the estmate of desrablty for that assgnment. The balance between copyng the fragments of the best soluton and generatng the rest usng preference values s controlled by the parameter λ, whch s the fracton of coped values among the total number of assgnments. Thu λ =0 corresponds to use of the RL algorthm to make all assgnments. Each cell had a probablty of beng coped equal to / n n 1 λ n λ and the remanng on average ( ) postons were flled usng acton-values Q l, f ) ( reflectng desrablty of assgnng faclty f j to locaton l. In the QAP, there s no obvous order n whch assgnments should be made. It makes the applcaton of bootstrappng RL algorthms such as Q-learnng dffcult unless some order s mposed that would put a strong bas on soluton generaton. There are a number of ways to resolve ths dffculty, but n the context of the present approach, a smple Monte Carlo update (4.2) that does not requre a partcular order was used, at the prce of slower convergence. The authors used a bootstrappng Q-learnng update rule n applcaton of ths approach to the Asymmetrc Travelng Salesman Problem (ATSP) descrbed n Magkkh and Punch (1999). In ths applcaton, the acton-value Q l, f ), were learned usng ( smple Monte Carlo update: Q l, f = Q l, f + α r Q l, f (4.2) ( ) ( ) ( ( ) where reward r was calculated on the bass of the average ftness of two parents accordng to (2.6). Generated solutons were mproved by a smple 1-Opt optmzer. 4.2 Bayesan Approach To mplement the approach capturng nterdependences drectly, one RL algorthm, a replca of the agent n the populaton from the populaton-based approach, was used. Ths RL algorthm was augmented wth an n 4 matrx of condtonal average rewards. The O(n 2 ) procedure (2.5) was used to update the entres of ths matrx. The acton-values were computed accordng to (2.4). The reward r for these two updates was calculated by (2.6), where avft was an average ftness of last M solutons produced. The process of feasble soluton generaton was dentcal to the RLagent approach wth a few changes: posteror acton-values computed by (2.5) were used nstead of acton-values n (4.2). The procedure for acceptng a new soluton was also relaxed n comparson to the populaton-based approach. The new soluton s accepted as a new center f the reward s greater than some constant threshold T. Snce the average ftness of soluton decreases durng the course of mnmzaton, ths acceptance rule resembles an annealng schedule. The same ε-greedy rule was used as found n the populaton-based approach, however a dfferent parameter ε was generated for each teraton to mprove the ablty of escapng local mnma. As n the frst approach, a smple 1- Opt optmzer was used to mprove newly created solutons. 5 Results The expermental runs of the RL agent approach were based on a populaton of 50 agent whch s relatvely small for a GA, but was enough to obtan good results usng ths scheme. Roulette wheel selecton was used. The parameter λ was randomly generated n range the [0.7,0.95] for each applcaton of RL crossover. A generaton-based approach wth a crossover rate 0.1 was used. The learnng and selecton greedness parameters were n ranges [0.05,0.15] for α, and [0.4,0.95] for ε, respectvely. These ranges of parameters were found n a

6 seres of prelmnary (not descrbed) experments. Each of the agents n the populaton was assgned a combnaton of parameters n these ranges durng ntalzaton. Thus there was a broad range of agent type based on a random selecton of the varous control parameters. For the approach based on Bayes rule, the parameters were β =0.1, α=0.1, T=0.55, M=300. The parameters ε and λ were randomly generated n the ntervals [0.3,0.99] and [0.5,0.9], respectvely, for each teraton. Results of RL agents and Bayesan approaches based on the averagng of 10 runs over some of the benchmark problems from QAPLIB by Burkard et al. (1997) are gven n Tables 1 and 2, respectvely (n Appendx). Snce the RL-agent approach has many features n common wth AS and GA, those approaches are used as a comparson. The columns AS and GA+LS n Table 1 show the results obtaned wth AS due to Manezzo and Colorn (1998) and GA wth local search by Merz and Fresleben (1997) respectvely. The RL algorthm wth Bayesan correcton was compared wth two other one-pont methods to contrast the qualty of search and the ablty to scale. The columns AS and GRASP n Table 2 show the results obtaned wth AS due to Manezzo and Colorn (1998) and greedy randomzed search procedure (GRASP) by L et al. (1994). The RL algorthm wth Bayesan correcton was tested on the same set of benchmarks as AS n Manezzo and Colorn (1998). Unfortunately, only the values of the best-found soluton are avalable for AS and GRASP. Thu we cannot compare them on average. However, the RL algorthm wth Bayesan correcton obtaned better results n terms of the best-found soluton n 7 out of 34 benchmarks. There was a te n the remanng 27 test problems. However, the presented RL algorthm does not scale well for the larger problems. In the case of the populaton-based approach, the results are much better. The RL-agents acheved the same or better performance on all test problems n comparson to ASbased algorthms. In comparson to GA+LS, the approach presented showed results whch were better on some benchmarks and slghtly worse on the others (of 15 problem 8 better and 7 worse). It can be concluded that the RL-agents approach and GA+LS were qute compettve. One of the remarkable features s the consstency of the search n the RL-agents approach: the presented algorthms found the optmum or best-known soluton n each of the 10 runs on all small and moderate-szed nstances. Ths s not the case wth GA+LS, whch had a non-zero standard devaton of the best-found solutons even on relatvely smple QAP benchmarks such as Nug30 or Kra30a. Unfortunately, only a small number of benchmark results for GA+LS are avalable, whch precludes more detaled comparson. Comparng the two approaches presented n ths paper, the populaton of RL agents certanly wns. The RL algorthm wth Bayesan correcton cannot provde a qualty of search equvalent to the frst approach. Even though our experments show some advantage to the Bayesan approach over other known non-populaton-based technque t does not compare well wth ether RL agents or GA. RL wth Bayesan correcton has an advantage n the number of functon evaluatons n comparson wth populaton-based approache though t has hgh space requrements. 6 Conclusons and Future Work The results of the approach usng a populaton of RL agents are compettve wth the other search technques. The authors also appled ths approach to ATSP and obtaned good results (Magkkh and Punch 1999). The approaches presented addresses the two major problems of RL algorthms n applcaton to optmzaton, namely, the local character of search and coarse state representaton. It has been shown that these problems can be overcome to obtan a global search technque capable of producng good results. There are stll many ssues to be addressed. One of them s to show that the preference values are a computatonally cheaper and more precse way of mantanng desrablty n comparson to GA and other search technque and f so, under what condtons? There are many other problem such as the absence of natural orderng n QAP and many other search problem whch can result n complcatons when bootstrappng RL update rules are used. The hgh space complexty of the Bayesan approach gves t more theoretcal rather than practcal value f a tabular representaton for the value functon s used. The authors are workng on mplementaton of ths approach usng functon approxmaton. However, n spte of all these and other dffculte the results obtaned are very encouragng. References Sutton, R. and Barto, A. (1997). Renforcement Learnng: An Introducton. MIT Press. Bellman, R. (1956). A Problem n Sequental Desgn of Experments, Sakhuya, 16: Dorgo, M. (1992). Optmzaton, Learnng and Natural Algorthms. Ph.D.Thes Poltecnco d Mlano, Italy, n Italan. Dorgo, M., Manezzo, V., and Colorn, A. (1996). The Ant System: Optmzaton by a Colony of Cooperatng Agent IEEE Trans. on SMC-Part B, 26 (1):29-41, IEEE Press.

7 Gambardella, L. and Dorgo, M. (1995). Ant-Q: A Renforcement Learnng Approach to the Travelng Salesman Problem. In Proc. 12th Int. Conf. on Machne Learnng, , Morgan Kaufmann. Crte R. and Barto, A. (1996). Improvng Elevator Performance usng Renforcement Learnng, Advances n Neural Informaton Processng Systems: Proc. of the 1999 Conf., , MIT Press. Sngh, S. and Bertseka D. (1996). Renforcement Learnng for Dynamc Channel Allocaton n Cellular Telephone Systems. In Proc. of Advances n Neural Informaton Processng System , MIT Press. Zhang W. and Detterch T. (1996). Hgh Performance Job-Shop Schedulng wth a Tme-delay TD(λ) Network. In Proc. of Advances n Neural Informaton Processng System , MIT Press. Holland, J. (1975). Adaptaton n Natural and Artfcal Systems. Unversty of Mchcan Press. Grefenstette, J. et al (1985). Genetc algorthms for the travelng salesman problem. In Proc. of 1 st Int. Conf. of Genetc Algorthms and ther applcaton , Lawrence Erlbaum Assocates Publshers. Boyan, J. and Moore A. (1998). "Learnng Evaluaton Functons for Global Optmzaton and Boolean Satsfablty", 15 Natonal Conf. on AI, AAAI. Magkkh, V. and Punch, W. (1999). An Approach to Solvng Combnatoral Optmzaton Problems Usng a Populaton of Renforcement Learnng Agents, To appear n Proc. of the Genetc and Evolutonary Computaton Conference (GECCO-99), Morgan Kaufmann. Burkard, R., Karsch, S., and Rendl, F. (1997). QAPLIB - A Quadratc Assgnment Problem Lbrary. Journal of Global Optmzaton, 10: Manezzo, V. and Colorn, A. (1998). The Ant System Appled to the Quadratc Assgnment Problem. To appear n IEEE transactons on Knowledge and Data Engneerng. Merz P. and Fresleben B. (1997a). A Genetc Local Search Approach to the Quadratc Assgnment Problem. In Proc. of the 7th Int. Conf. on GA (ICGA'97), Appendx Table 1: Results of RL-Agent approach on QAPLIB by Burkard et al. (1997). The meanng of the columns s as follows: Benchmark the name of the benchmark; Opt./BKS. optmal or best known soluton for ths problem; Best the best result found by the populaton of RL agents n 10 runs; Average average among the best solutons found n 10 runs; Std. Dev. standard devaton of the dstrbuton of the values of the best solutons found; NFE average number of the functon evaluatons to fnd the best soluton; AS - the ftness of the best soluton obtaned by the AS by Manezzo and Colorn (1998); GA+LS the average ftness of soluton obtaned by GA wth local search as descrbed n P. Merz, B. Fresleben (1997). The best soluton among the three technques s bolded. Benchmark Opt./BKS. Best Average Std. Dev NFE AS GA+LS Bur26a N/A Bur26b N/A Bur26c N/A Bur26d N/A Bur26e N/A Bur26f N/A Bur26g N/A Chr20a N/A Chr20b N/A Chr20c N/A Chr22a N/A Chr22b N/A Esc32a N/A Kra30a N/A Kra30b N/A Lpa20a N/A Lpa30a N/A Lpa40a N/A Nug N/A Nug Scr N/A Ste36a Ste36b N/A Ste36c N/A Sko100a N/A Ta60a N/A

8 Benchmark Opt./BKS. Best Average Std. Dev NFE AS GA+LS Ta60b N/A Tal100a N/A Tal100b N/A Ta150b N/A Ta256c N/A Tho N/A Tho N/A Tho N/A Table 2: Results of RL algorthm wth Bayesan correcton on QAPLIB by Burkard et al. (1997). The meanng of the columns s as follows: Benchmark the name of the benchmark; Opt./BKS. optmal or best known soluton for ths problem; Best the best result found by the populaton of RL agents n 10 runs; Average average among the best solutons found n 10 runs; Std. Dev. standard devaton of the dstrbuton of the values of the best solutons found; NFE average number of the functon evaluatons to fnd the best soluton; AS - the ftness of the best soluton obtaned by the AS by Manezzo and Colorn (1998); GRASP the best ftness of soluton obtaned by greedy randomzed search procedure (GRASP) by L, Pardalos and Resende (1994) as gven n Manezzo and Colorn (1998). The best soluton among the three technques s bolded. Benchmark Opt./BKS. Best Average Std. Dev. NFE AS GRASP Bur26a Bur26b Bur26c Bur26d Bur26e Bur26f Bur26g Chr20a Chr20b Chr20c Chr22a Chr22b Els N/A N/A Esc32a Esc32b Esc32c Esc32d Esc32e Esc32f Esc32g Esc64a N/A N/A Kra30a Kra30b Lpa20a Lpa30a Lpa40a Nug Nug Scr Ste36a Ste36b Ste36c Tho Tho Sko N/A N/A

benefit is 2, paid if the policyholder dies within the year, and probability of death within the year is ).

benefit is 2, paid if the policyholder dies within the year, and probability of death within the year is ). REVIEW OF RISK MANAGEMENT CONCEPTS LOSS DISTRIBUTIONS AND INSURANCE Loss and nsurance: When someone s subject to the rsk of ncurrng a fnancal loss, the loss s generally modeled usng a random varable or

More information

What is Candidate Sampling

What is Candidate Sampling What s Canddate Samplng Say we have a multclass or mult label problem where each tranng example ( x, T ) conssts of a context x a small (mult)set of target classes T out of a large unverse L of possble

More information

PSYCHOLOGICAL RESEARCH (PYC 304-C) Lecture 12

PSYCHOLOGICAL RESEARCH (PYC 304-C) Lecture 12 14 The Ch-squared dstrbuton PSYCHOLOGICAL RESEARCH (PYC 304-C) Lecture 1 If a normal varable X, havng mean µ and varance σ, s standardsed, the new varable Z has a mean 0 and varance 1. When ths standardsed

More information

Module 2 LOSSLESS IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur

Module 2 LOSSLESS IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur Module LOSSLESS IMAGE COMPRESSION SYSTEMS Lesson 3 Lossless Compresson: Huffman Codng Instructonal Objectves At the end of ths lesson, the students should be able to:. Defne and measure source entropy..

More information

Ants Can Schedule Software Projects

Ants Can Schedule Software Projects Ants Can Schedule Software Proects Broderck Crawford 1,2, Rcardo Soto 1,3, Frankln Johnson 4, and Erc Monfroy 5 1 Pontfca Unversdad Católca de Valparaíso, Chle FrstName.Name@ucv.cl 2 Unversdad Fns Terrae,

More information

The Development of Web Log Mining Based on Improve-K-Means Clustering Analysis

The Development of Web Log Mining Based on Improve-K-Means Clustering Analysis The Development of Web Log Mnng Based on Improve-K-Means Clusterng Analyss TngZhong Wang * College of Informaton Technology, Luoyang Normal Unversty, Luoyang, 471022, Chna wangtngzhong2@sna.cn Abstract.

More information

The OC Curve of Attribute Acceptance Plans

The OC Curve of Attribute Acceptance Plans The OC Curve of Attrbute Acceptance Plans The Operatng Characterstc (OC) curve descrbes the probablty of acceptng a lot as a functon of the lot s qualty. Fgure 1 shows a typcal OC Curve. 10 8 6 4 1 3 4

More information

On the Optimal Control of a Cascade of Hydro-Electric Power Stations

On the Optimal Control of a Cascade of Hydro-Electric Power Stations On the Optmal Control of a Cascade of Hydro-Electrc Power Statons M.C.M. Guedes a, A.F. Rbero a, G.V. Smrnov b and S. Vlela c a Department of Mathematcs, School of Scences, Unversty of Porto, Portugal;

More information

The Greedy Method. Introduction. 0/1 Knapsack Problem

The Greedy Method. Introduction. 0/1 Knapsack Problem The Greedy Method Introducton We have completed data structures. We now are gong to look at algorthm desgn methods. Often we are lookng at optmzaton problems whose performance s exponental. For an optmzaton

More information

Recurrence. 1 Definitions and main statements

Recurrence. 1 Definitions and main statements Recurrence 1 Defntons and man statements Let X n, n = 0, 1, 2,... be a MC wth the state space S = (1, 2,...), transton probabltes p j = P {X n+1 = j X n = }, and the transton matrx P = (p j ),j S def.

More information

An Alternative Way to Measure Private Equity Performance

An Alternative Way to Measure Private Equity Performance An Alternatve Way to Measure Prvate Equty Performance Peter Todd Parlux Investment Technology LLC Summary Internal Rate of Return (IRR) s probably the most common way to measure the performance of prvate

More information

J. Parallel Distrib. Comput.

J. Parallel Distrib. Comput. J. Parallel Dstrb. Comput. 71 (2011) 62 76 Contents lsts avalable at ScenceDrect J. Parallel Dstrb. Comput. journal homepage: www.elsever.com/locate/jpdc Optmzng server placement n dstrbuted systems n

More information

Institute of Informatics, Faculty of Business and Management, Brno University of Technology,Czech Republic

Institute of Informatics, Faculty of Business and Management, Brno University of Technology,Czech Republic Lagrange Multplers as Quanttatve Indcators n Economcs Ivan Mezník Insttute of Informatcs, Faculty of Busness and Management, Brno Unversty of TechnologCzech Republc Abstract The quanttatve role of Lagrange

More information

Software project management with GAs

Software project management with GAs Informaton Scences 177 (27) 238 241 www.elsever.com/locate/ns Software project management wth GAs Enrque Alba *, J. Francsco Chcano Unversty of Málaga, Grupo GISUM, Departamento de Lenguajes y Cencas de

More information

Ant Colony Optimization for Economic Generator Scheduling and Load Dispatch

Ant Colony Optimization for Economic Generator Scheduling and Load Dispatch Proceedngs of the th WSEAS Int. Conf. on EVOLUTIONARY COMPUTING, Lsbon, Portugal, June 1-18, 5 (pp17-175) Ant Colony Optmzaton for Economc Generator Schedulng and Load Dspatch K. S. Swarup Abstract Feasblty

More information

A New Task Scheduling Algorithm Based on Improved Genetic Algorithm

A New Task Scheduling Algorithm Based on Improved Genetic Algorithm A New Task Schedulng Algorthm Based on Improved Genetc Algorthm n Cloud Computng Envronment Congcong Xong, Long Feng, Lxan Chen A New Task Schedulng Algorthm Based on Improved Genetc Algorthm n Cloud Computng

More information

DEFINING %COMPLETE IN MICROSOFT PROJECT

DEFINING %COMPLETE IN MICROSOFT PROJECT CelersSystems DEFINING %COMPLETE IN MICROSOFT PROJECT PREPARED BY James E Aksel, PMP, PMI-SP, MVP For Addtonal Informaton about Earned Value Management Systems and reportng, please contact: CelersSystems,

More information

CS 2750 Machine Learning. Lecture 3. Density estimation. CS 2750 Machine Learning. Announcements

CS 2750 Machine Learning. Lecture 3. Density estimation. CS 2750 Machine Learning. Announcements Lecture 3 Densty estmaton Mlos Hauskrecht mlos@cs.ptt.edu 5329 Sennott Square Next lecture: Matlab tutoral Announcements Rules for attendng the class: Regstered for credt Regstered for audt (only f there

More information

Number of Levels Cumulative Annual operating Income per year construction costs costs ($) ($) ($) 1 600,000 35,000 100,000 2 2,200,000 60,000 350,000

Number of Levels Cumulative Annual operating Income per year construction costs costs ($) ($) ($) 1 600,000 35,000 100,000 2 2,200,000 60,000 350,000 Problem Set 5 Solutons 1 MIT s consderng buldng a new car park near Kendall Square. o unversty funds are avalable (overhead rates are under pressure and the new faclty would have to pay for tself from

More information

"Research Note" APPLICATION OF CHARGE SIMULATION METHOD TO ELECTRIC FIELD CALCULATION IN THE POWER CABLES *

Research Note APPLICATION OF CHARGE SIMULATION METHOD TO ELECTRIC FIELD CALCULATION IN THE POWER CABLES * Iranan Journal of Scence & Technology, Transacton B, Engneerng, ol. 30, No. B6, 789-794 rnted n The Islamc Republc of Iran, 006 Shraz Unversty "Research Note" ALICATION OF CHARGE SIMULATION METHOD TO ELECTRIC

More information

Luby s Alg. for Maximal Independent Sets using Pairwise Independence

Luby s Alg. for Maximal Independent Sets using Pairwise Independence Lecture Notes for Randomzed Algorthms Luby s Alg. for Maxmal Independent Sets usng Parwse Independence Last Updated by Erc Vgoda on February, 006 8. Maxmal Independent Sets For a graph G = (V, E), an ndependent

More information

A Probabilistic Theory of Coherence

A Probabilistic Theory of Coherence A Probablstc Theory of Coherence BRANDEN FITELSON. The Coherence Measure C Let E be a set of n propostons E,..., E n. We seek a probablstc measure C(E) of the degree of coherence of E. Intutvely, we want

More information

Answer: A). There is a flatter IS curve in the high MPC economy. Original LM LM after increase in M. IS curve for low MPC economy

Answer: A). There is a flatter IS curve in the high MPC economy. Original LM LM after increase in M. IS curve for low MPC economy 4.02 Quz Solutons Fall 2004 Multple-Choce Questons (30/00 ponts) Please, crcle the correct answer for each of the followng 0 multple-choce questons. For each queston, only one of the answers s correct.

More information

Can Auto Liability Insurance Purchases Signal Risk Attitude?

Can Auto Liability Insurance Purchases Signal Risk Attitude? Internatonal Journal of Busness and Economcs, 2011, Vol. 10, No. 2, 159-164 Can Auto Lablty Insurance Purchases Sgnal Rsk Atttude? Chu-Shu L Department of Internatonal Busness, Asa Unversty, Tawan Sheng-Chang

More information

Mooring Pattern Optimization using Genetic Algorithms

Mooring Pattern Optimization using Genetic Algorithms 6th World Congresses of Structural and Multdscplnary Optmzaton Ro de Janero, 30 May - 03 June 005, Brazl Moorng Pattern Optmzaton usng Genetc Algorthms Alonso J. Juvnao Carbono, Ivan F. M. Menezes Luz

More information

8.5 UNITARY AND HERMITIAN MATRICES. The conjugate transpose of a complex matrix A, denoted by A*, is given by

8.5 UNITARY AND HERMITIAN MATRICES. The conjugate transpose of a complex matrix A, denoted by A*, is given by 6 CHAPTER 8 COMPLEX VECTOR SPACES 5. Fnd the kernel of the lnear transformaton gven n Exercse 5. In Exercses 55 and 56, fnd the mage of v, for the ndcated composton, where and are gven by the followng

More information

Implementation of Deutsch's Algorithm Using Mathcad

Implementation of Deutsch's Algorithm Using Mathcad Implementaton of Deutsch's Algorthm Usng Mathcad Frank Roux The followng s a Mathcad mplementaton of Davd Deutsch's quantum computer prototype as presented on pages - n "Machnes, Logc and Quantum Physcs"

More information

SCHEDULING OF CONSTRUCTION PROJECTS BY MEANS OF EVOLUTIONARY ALGORITHMS

SCHEDULING OF CONSTRUCTION PROJECTS BY MEANS OF EVOLUTIONARY ALGORITHMS SCHEDULING OF CONSTRUCTION PROJECTS BY MEANS OF EVOLUTIONARY ALGORITHMS Magdalena Rogalska 1, Wocech Bożeko 2,Zdzsław Heduck 3, 1 Lubln Unversty of Technology, 2- Lubln, Nadbystrzycka 4., Poland. E-mal:rogalska@akropols.pol.lubln.pl

More information

Data Broadcast on a Multi-System Heterogeneous Overlayed Wireless Network *

Data Broadcast on a Multi-System Heterogeneous Overlayed Wireless Network * JOURNAL OF INFORMATION SCIENCE AND ENGINEERING 24, 819-840 (2008) Data Broadcast on a Mult-System Heterogeneous Overlayed Wreless Network * Department of Computer Scence Natonal Chao Tung Unversty Hsnchu,

More information

How Sets of Coherent Probabilities May Serve as Models for Degrees of Incoherence

How Sets of Coherent Probabilities May Serve as Models for Degrees of Incoherence 1 st Internatonal Symposum on Imprecse Probabltes and Ther Applcatons, Ghent, Belgum, 29 June 2 July 1999 How Sets of Coherent Probabltes May Serve as Models for Degrees of Incoherence Mar J. Schervsh

More information

Credit Limit Optimization (CLO) for Credit Cards

Credit Limit Optimization (CLO) for Credit Cards Credt Lmt Optmzaton (CLO) for Credt Cards Vay S. Desa CSCC IX, Ednburgh September 8, 2005 Copyrght 2003, SAS Insttute Inc. All rghts reserved. SAS Propretary Agenda Background Tradtonal approaches to credt

More information

Joint Scheduling of Processing and Shuffle Phases in MapReduce Systems

Joint Scheduling of Processing and Shuffle Phases in MapReduce Systems Jont Schedulng of Processng and Shuffle Phases n MapReduce Systems Fangfe Chen, Mural Kodalam, T. V. Lakshman Department of Computer Scence and Engneerng, The Penn State Unversty Bell Laboratores, Alcatel-Lucent

More information

An ACO Algorithm for. the Graph Coloring Problem

An ACO Algorithm for. the Graph Coloring Problem Int. J. Contemp. Math. Scences, Vol. 3, 2008, no. 6, 293-304 An ACO Algorthm for the Graph Colorng Problem Ehsan Salar and Kourosh Eshgh Department of Industral Engneerng Sharf Unversty of Technology,

More information

Logistic Regression. Lecture 4: More classifiers and classes. Logistic regression. Adaboost. Optimization. Multiple class classification

Logistic Regression. Lecture 4: More classifiers and classes. Logistic regression. Adaboost. Optimization. Multiple class classification Lecture 4: More classfers and classes C4B Machne Learnng Hlary 20 A. Zsserman Logstc regresson Loss functons revsted Adaboost Loss functons revsted Optmzaton Multple class classfcaton Logstc Regresson

More information

Blending Roulette Wheel Selection & Rank Selection in Genetic Algorithms

Blending Roulette Wheel Selection & Rank Selection in Genetic Algorithms Internatonal Journal of Machne Learnng and Computng, Vol. 2, o. 4, August 2012 Blendng Roulette Wheel Selecton & Rank Selecton n Genetc Algorthms Rakesh Kumar, Senor Member, IACSIT and Jyotshree, Member,

More information

SUPPLIER FINANCING AND STOCK MANAGEMENT. A JOINT VIEW.

SUPPLIER FINANCING AND STOCK MANAGEMENT. A JOINT VIEW. SUPPLIER FINANCING AND STOCK MANAGEMENT. A JOINT VIEW. Lucía Isabel García Cebrán Departamento de Economía y Dreccón de Empresas Unversdad de Zaragoza Gran Vía, 2 50.005 Zaragoza (Span) Phone: 976-76-10-00

More information

Extending Probabilistic Dynamic Epistemic Logic

Extending Probabilistic Dynamic Epistemic Logic Extendng Probablstc Dynamc Epstemc Logc Joshua Sack May 29, 2008 Probablty Space Defnton A probablty space s a tuple (S, A, µ), where 1 S s a set called the sample space. 2 A P(S) s a σ-algebra: a set

More information

Feature selection for intrusion detection. Slobodan Petrović NISlab, Gjøvik University College

Feature selection for intrusion detection. Slobodan Petrović NISlab, Gjøvik University College Feature selecton for ntruson detecton Slobodan Petrovć NISlab, Gjøvk Unversty College Contents The feature selecton problem Intruson detecton Traffc features relevant for IDS The CFS measure The mrmr measure

More information

An Analysis of Dynamic Severity and Population Size

An Analysis of Dynamic Severity and Population Size An Analyss of Dynamc Severty and Populaton Sze Karsten Wecker Unversty of Stuttgart, Insttute of Computer Scence, Bretwesenstr. 2 22, 7565 Stuttgart, Germany, emal: Karsten.Wecker@nformatk.un-stuttgart.de

More information

Improved SVM in Cloud Computing Information Mining

Improved SVM in Cloud Computing Information Mining Internatonal Journal of Grd Dstrbuton Computng Vol.8, No.1 (015), pp.33-40 http://dx.do.org/10.1457/jgdc.015.8.1.04 Improved n Cloud Computng Informaton Mnng Lvshuhong (ZhengDe polytechnc college JangSu

More information

Learning the Best K-th Channel for QoS Provisioning in Cognitive Networks

Learning the Best K-th Channel for QoS Provisioning in Cognitive Networks 000 001 002 003 004 005 006 007 008 009 010 011 012 013 014 015 016 017 018 019 020 021 022 023 024 025 026 027 028 029 030 031 032 033 034 035 036 037 038 039 040 041 042 043 044 045 046 047 048 049 050

More information

A hybrid global optimization algorithm based on parallel chaos optimization and outlook algorithm

A hybrid global optimization algorithm based on parallel chaos optimization and outlook algorithm Avalable onlne www.ocpr.com Journal of Chemcal and Pharmaceutcal Research, 2014, 6(7):1884-1889 Research Artcle ISSN : 0975-7384 CODEN(USA) : JCPRC5 A hybrd global optmzaton algorthm based on parallel

More information

Using Multi-objective Metaheuristics to Solve the Software Project Scheduling Problem

Using Multi-objective Metaheuristics to Solve the Software Project Scheduling Problem Usng Mult-obectve Metaheurstcs to Solve the Software Proect Schedulng Problem Francsco Chcano Unversty of Málaga, Span chcano@lcc.uma.es Francsco Luna Unversty of Málaga, Span flv@lcc.uma.es Enrque Alba

More information

Optimization Model of Reliable Data Storage in Cloud Environment Using Genetic Algorithm

Optimization Model of Reliable Data Storage in Cloud Environment Using Genetic Algorithm Internatonal Journal of Grd Dstrbuton Computng, pp.175-190 http://dx.do.org/10.14257/gdc.2014.7.6.14 Optmzaton odel of Relable Data Storage n Cloud Envronment Usng Genetc Algorthm Feng Lu 1,2,3, Hatao

More information

L10: Linear discriminants analysis

L10: Linear discriminants analysis L0: Lnear dscrmnants analyss Lnear dscrmnant analyss, two classes Lnear dscrmnant analyss, C classes LDA vs. PCA Lmtatons of LDA Varants of LDA Other dmensonalty reducton methods CSCE 666 Pattern Analyss

More information

Traffic-light a stress test for life insurance provisions

Traffic-light a stress test for life insurance provisions MEMORANDUM Date 006-09-7 Authors Bengt von Bahr, Göran Ronge Traffc-lght a stress test for lfe nsurance provsons Fnansnspetonen P.O. Box 6750 SE-113 85 Stocholm [Sveavägen 167] Tel +46 8 787 80 00 Fax

More information

Dynamic Pricing for Smart Grid with Reinforcement Learning

Dynamic Pricing for Smart Grid with Reinforcement Learning Dynamc Prcng for Smart Grd wth Renforcement Learnng Byung-Gook Km, Yu Zhang, Mhaela van der Schaar, and Jang-Won Lee Samsung Electroncs, Suwon, Korea Department of Electrcal Engneerng, UCLA, Los Angeles,

More information

Forecasting the Demand of Emergency Supplies: Based on the CBR Theory and BP Neural Network

Forecasting the Demand of Emergency Supplies: Based on the CBR Theory and BP Neural Network 700 Proceedngs of the 8th Internatonal Conference on Innovaton & Management Forecastng the Demand of Emergency Supples: Based on the CBR Theory and BP Neural Network Fu Deqang, Lu Yun, L Changbng School

More information

Brigid Mullany, Ph.D University of North Carolina, Charlotte

Brigid Mullany, Ph.D University of North Carolina, Charlotte Evaluaton And Comparson Of The Dfferent Standards Used To Defne The Postonal Accuracy And Repeatablty Of Numercally Controlled Machnng Center Axes Brgd Mullany, Ph.D Unversty of North Carolna, Charlotte

More information

A DYNAMIC CRASHING METHOD FOR PROJECT MANAGEMENT USING SIMULATION-BASED OPTIMIZATION. Michael E. Kuhl Radhamés A. Tolentino-Peña

A DYNAMIC CRASHING METHOD FOR PROJECT MANAGEMENT USING SIMULATION-BASED OPTIMIZATION. Michael E. Kuhl Radhamés A. Tolentino-Peña Proceedngs of the 2008 Wnter Smulaton Conference S. J. Mason, R. R. Hll, L. Mönch, O. Rose, T. Jefferson, J. W. Fowler eds. A DYNAMIC CRASHING METHOD FOR PROJECT MANAGEMENT USING SIMULATION-BASED OPTIMIZATION

More information

IMPACT ANALYSIS OF A CELLULAR PHONE

IMPACT ANALYSIS OF A CELLULAR PHONE 4 th ASA & μeta Internatonal Conference IMPACT AALYSIS OF A CELLULAR PHOE We Lu, 2 Hongy L Bejng FEAonlne Engneerng Co.,Ltd. Bejng, Chna ABSTRACT Drop test smulaton plays an mportant role n nvestgatng

More information

Realistic Image Synthesis

Realistic Image Synthesis Realstc Image Synthess - Combned Samplng and Path Tracng - Phlpp Slusallek Karol Myszkowsk Vncent Pegoraro Overvew: Today Combned Samplng (Multple Importance Samplng) Renderng and Measurng Equaton Random

More information

Risk Model of Long-Term Production Scheduling in Open Pit Gold Mining

Risk Model of Long-Term Production Scheduling in Open Pit Gold Mining Rsk Model of Long-Term Producton Schedulng n Open Pt Gold Mnng R Halatchev 1 and P Lever 2 ABSTRACT Open pt gold mnng s an mportant sector of the Australan mnng ndustry. It uses large amounts of nvestments,

More information

Examensarbete. Rotating Workforce Scheduling. Caroline Granfeldt

Examensarbete. Rotating Workforce Scheduling. Caroline Granfeldt Examensarbete Rotatng Workforce Schedulng Carolne Granfeldt LTH - MAT - EX - - 2015 / 08 - - SE Rotatng Workforce Schedulng Optmerngslära, Lnköpngs Unverstet Carolne Granfeldt LTH - MAT - EX - - 2015

More information

Vision Mouse. Saurabh Sarkar a* University of Cincinnati, Cincinnati, USA ABSTRACT 1. INTRODUCTION

Vision Mouse. Saurabh Sarkar a* University of Cincinnati, Cincinnati, USA ABSTRACT 1. INTRODUCTION Vson Mouse Saurabh Sarkar a* a Unversty of Cncnnat, Cncnnat, USA ABSTRACT The report dscusses a vson based approach towards trackng of eyes and fngers. The report descrbes the process of locatng the possble

More information

Maintenance Scheduling by using the Bi-Criterion Algorithm of Preferential Anti-Pheromone

Maintenance Scheduling by using the Bi-Criterion Algorithm of Preferential Anti-Pheromone Leonardo ournal of Scences ISSN 583-0233 Issue 2, anuary-une 2008 p. 43-64 Mantenance Schedulng by usng the B-Crteron Algorthm of Preferental Ant-Pheromone Trantafyllos MYTAKIDIS and Arstds VLACHOS Department

More information

Formulating & Solving Integer Problems Chapter 11 289

Formulating & Solving Integer Problems Chapter 11 289 Formulatng & Solvng Integer Problems Chapter 11 289 The Optonal Stop TSP If we drop the requrement that every stop must be vsted, we then get the optonal stop TSP. Ths mght correspond to a ob sequencng

More information

Frequency Selective IQ Phase and IQ Amplitude Imbalance Adjustments for OFDM Direct Conversion Transmitters

Frequency Selective IQ Phase and IQ Amplitude Imbalance Adjustments for OFDM Direct Conversion Transmitters Frequency Selectve IQ Phase and IQ Ampltude Imbalance Adjustments for OFDM Drect Converson ransmtters Edmund Coersmeer, Ernst Zelnsk Noka, Meesmannstrasse 103, 44807 Bochum, Germany edmund.coersmeer@noka.com,

More information

Descriptive Models. Cluster Analysis. Example. General Applications of Clustering. Examples of Clustering Applications

Descriptive Models. Cluster Analysis. Example. General Applications of Clustering. Examples of Clustering Applications CMSC828G Prncples of Data Mnng Lecture #9 Today s Readng: HMS, chapter 9 Today s Lecture: Descrptve Modelng Clusterng Algorthms Descrptve Models model presents the man features of the data, a global summary

More information

Efficient Reinforcement Learning in Factored MDPs

Efficient Reinforcement Learning in Factored MDPs Effcent Renforcement Learnng n Factored MDPs Mchael Kearns AT&T Labs mkearns@research.att.com Daphne Koller Stanford Unversty koller@cs.stanford.edu Abstract We present a provably effcent and near-optmal

More information

ANALYZING THE RELATIONSHIPS BETWEEN QUALITY, TIME, AND COST IN PROJECT MANAGEMENT DECISION MAKING

ANALYZING THE RELATIONSHIPS BETWEEN QUALITY, TIME, AND COST IN PROJECT MANAGEMENT DECISION MAKING ANALYZING THE RELATIONSHIPS BETWEEN QUALITY, TIME, AND COST IN PROJECT MANAGEMENT DECISION MAKING Matthew J. Lberatore, Department of Management and Operatons, Vllanova Unversty, Vllanova, PA 19085, 610-519-4390,

More information

Support Vector Machines

Support Vector Machines Support Vector Machnes Max Wellng Department of Computer Scence Unversty of Toronto 10 Kng s College Road Toronto, M5S 3G5 Canada wellng@cs.toronto.edu Abstract Ths s a note to explan support vector machnes.

More information

) of the Cell class is created containing information about events associated with the cell. Events are added to the Cell instance

) of the Cell class is created containing information about events associated with the cell. Events are added to the Cell instance Calbraton Method Instances of the Cell class (one nstance for each FMS cell) contan ADC raw data and methods assocated wth each partcular FMS cell. The calbraton method ncludes event selecton (Class Cell

More information

To Fill or not to Fill: The Gas Station Problem

To Fill or not to Fill: The Gas Station Problem To Fll or not to Fll: The Gas Staton Problem Samr Khuller Azarakhsh Malekan Julán Mestre Abstract In ths paper we study several routng problems that generalze shortest paths and the Travelng Salesman Problem.

More information

Enabling P2P One-view Multi-party Video Conferencing

Enabling P2P One-view Multi-party Video Conferencing Enablng P2P One-vew Mult-party Vdeo Conferencng Yongxang Zhao, Yong Lu, Changja Chen, and JanYn Zhang Abstract Mult-Party Vdeo Conferencng (MPVC) facltates realtme group nteracton between users. Whle P2P

More information

An Interest-Oriented Network Evolution Mechanism for Online Communities

An Interest-Oriented Network Evolution Mechanism for Online Communities An Interest-Orented Network Evoluton Mechansm for Onlne Communtes Cahong Sun and Xaopng Yang School of Informaton, Renmn Unversty of Chna, Bejng 100872, P.R. Chna {chsun,yang}@ruc.edu.cn Abstract. Onlne

More information

Lecture 2: Single Layer Perceptrons Kevin Swingler

Lecture 2: Single Layer Perceptrons Kevin Swingler Lecture 2: Sngle Layer Perceptrons Kevn Sngler kms@cs.str.ac.uk Recap: McCulloch-Ptts Neuron Ths vastly smplfed model of real neurons s also knon as a Threshold Logc Unt: W 2 A Y 3 n W n. A set of synapses

More information

Course outline. Financial Time Series Analysis. Overview. Data analysis. Predictive signal. Trading strategy

Course outline. Financial Time Series Analysis. Overview. Data analysis. Predictive signal. Trading strategy Fnancal Tme Seres Analyss Patrck McSharry patrck@mcsharry.net www.mcsharry.net Trnty Term 2014 Mathematcal Insttute Unversty of Oxford Course outlne 1. Data analyss, probablty, correlatons, vsualsaton

More information

A Performance Analysis of View Maintenance Techniques for Data Warehouses

A Performance Analysis of View Maintenance Techniques for Data Warehouses A Performance Analyss of Vew Mantenance Technques for Data Warehouses Xng Wang Dell Computer Corporaton Round Roc, Texas Le Gruenwald The nversty of Olahoma School of Computer Scence orman, OK 739 Guangtao

More information

An MILP model for planning of batch plants operating in a campaign-mode

An MILP model for planning of batch plants operating in a campaign-mode An MILP model for plannng of batch plants operatng n a campagn-mode Yanna Fumero Insttuto de Desarrollo y Dseño CONICET UTN yfumero@santafe-concet.gov.ar Gabrela Corsano Insttuto de Desarrollo y Dseño

More information

Statistical algorithms in Review Manager 5

Statistical algorithms in Review Manager 5 Statstcal algorthms n Reve Manager 5 Jonathan J Deeks and Julan PT Hggns on behalf of the Statstcal Methods Group of The Cochrane Collaboraton August 00 Data structure Consder a meta-analyss of k studes

More information

行 政 院 國 家 科 學 委 員 會 補 助 專 題 研 究 計 畫 成 果 報 告 期 中 進 度 報 告

行 政 院 國 家 科 學 委 員 會 補 助 專 題 研 究 計 畫 成 果 報 告 期 中 進 度 報 告 行 政 院 國 家 科 學 委 員 會 補 助 專 題 研 究 計 畫 成 果 報 告 期 中 進 度 報 告 畫 類 別 : 個 別 型 計 畫 半 導 體 產 業 大 型 廠 房 之 設 施 規 劃 計 畫 編 號 :NSC 96-2628-E-009-026-MY3 執 行 期 間 : 2007 年 8 月 1 日 至 2010 年 7 月 31 日 計 畫 主 持 人 : 巫 木 誠 共 同

More information

Calculation of Sampling Weights

Calculation of Sampling Weights Perre Foy Statstcs Canada 4 Calculaton of Samplng Weghts 4.1 OVERVIEW The basc sample desgn used n TIMSS Populatons 1 and 2 was a two-stage stratfed cluster desgn. 1 The frst stage conssted of a sample

More information

1. Fundamentals of probability theory 2. Emergence of communication traffic 3. Stochastic & Markovian Processes (SP & MP)

1. Fundamentals of probability theory 2. Emergence of communication traffic 3. Stochastic & Markovian Processes (SP & MP) 6.3 / -- Communcaton Networks II (Görg) SS20 -- www.comnets.un-bremen.de Communcaton Networks II Contents. Fundamentals of probablty theory 2. Emergence of communcaton traffc 3. Stochastc & Markovan Processes

More information

Traffic State Estimation in the Traffic Management Center of Berlin

Traffic State Estimation in the Traffic Management Center of Berlin Traffc State Estmaton n the Traffc Management Center of Berln Authors: Peter Vortsch, PTV AG, Stumpfstrasse, D-763 Karlsruhe, Germany phone ++49/72/965/35, emal peter.vortsch@ptv.de Peter Möhl, PTV AG,

More information

The Current Employment Statistics (CES) survey,

The Current Employment Statistics (CES) survey, Busness Brths and Deaths Impact of busness brths and deaths n the payroll survey The CES probablty-based sample redesgn accounts for most busness brth employment through the mputaton of busness deaths,

More information

How To Solve A Problem In A Powerline (Powerline) With A Powerbook (Powerbook)

How To Solve A Problem In A Powerline (Powerline) With A Powerbook (Powerbook) MIT 8.996: Topc n TCS: Internet Research Problems Sprng 2002 Lecture 7 March 20, 2002 Lecturer: Bran Dean Global Load Balancng Scrbe: John Kogel, Ben Leong In today s lecture, we dscuss global load balancng

More information

How To Solve An Onlne Control Polcy On A Vrtualzed Data Center

How To Solve An Onlne Control Polcy On A Vrtualzed Data Center Dynamc Resource Allocaton and Power Management n Vrtualzed Data Centers Rahul Urgaonkar, Ulas C. Kozat, Ken Igarash, Mchael J. Neely urgaonka@usc.edu, {kozat, garash}@docomolabs-usa.com, mjneely@usc.edu

More information

14.74 Lecture 5: Health (2)

14.74 Lecture 5: Health (2) 14.74 Lecture 5: Health (2) Esther Duflo February 17, 2004 1 Possble Interventons Last tme we dscussed possble nterventons. Let s take one: provdng ron supplements to people, for example. From the data,

More information

A Load-Balancing Algorithm for Cluster-based Multi-core Web Servers

A Load-Balancing Algorithm for Cluster-based Multi-core Web Servers Journal of Computatonal Informaton Systems 7: 13 (2011) 4740-4747 Avalable at http://www.jofcs.com A Load-Balancng Algorthm for Cluster-based Mult-core Web Servers Guohua YOU, Yng ZHAO College of Informaton

More information

CHOLESTEROL REFERENCE METHOD LABORATORY NETWORK. Sample Stability Protocol

CHOLESTEROL REFERENCE METHOD LABORATORY NETWORK. Sample Stability Protocol CHOLESTEROL REFERENCE METHOD LABORATORY NETWORK Sample Stablty Protocol Background The Cholesterol Reference Method Laboratory Network (CRMLN) developed certfcaton protocols for total cholesterol, HDL

More information

1 Example 1: Axis-aligned rectangles

1 Example 1: Axis-aligned rectangles COS 511: Theoretcal Machne Learnng Lecturer: Rob Schapre Lecture # 6 Scrbe: Aaron Schld February 21, 2013 Last class, we dscussed an analogue for Occam s Razor for nfnte hypothess spaces that, n conjuncton

More information

AN APPOINTMENT ORDER OUTPATIENT SCHEDULING SYSTEM THAT IMPROVES OUTPATIENT EXPERIENCE

AN APPOINTMENT ORDER OUTPATIENT SCHEDULING SYSTEM THAT IMPROVES OUTPATIENT EXPERIENCE AN APPOINTMENT ORDER OUTPATIENT SCHEDULING SYSTEM THAT IMPROVES OUTPATIENT EXPERIENCE Yu-L Huang Industral Engneerng Department New Mexco State Unversty Las Cruces, New Mexco 88003, U.S.A. Abstract Patent

More information

Abstract. Clustering ensembles have emerged as a powerful method for improving both the

Abstract. Clustering ensembles have emerged as a powerful method for improving both the Clusterng Ensembles: {topchyal, Models jan, of punch}@cse.msu.edu Consensus and Weak Parttons * Alexander Topchy, Anl K. Jan, and Wllam Punch Department of Computer Scence and Engneerng, Mchgan State Unversty

More information

APPLICATION OF PROBE DATA COLLECTED VIA INFRARED BEACONS TO TRAFFIC MANEGEMENT

APPLICATION OF PROBE DATA COLLECTED VIA INFRARED BEACONS TO TRAFFIC MANEGEMENT APPLICATION OF PROBE DATA COLLECTED VIA INFRARED BEACONS TO TRAFFIC MANEGEMENT Toshhko Oda (1), Kochro Iwaoka (2) (1), (2) Infrastructure Systems Busness Unt, Panasonc System Networks Co., Ltd. Saedo-cho

More information

PRIVATE SCHOOL CHOICE: THE EFFECTS OF RELIGIOUS AFFILIATION AND PARTICIPATION

PRIVATE SCHOOL CHOICE: THE EFFECTS OF RELIGIOUS AFFILIATION AND PARTICIPATION PRIVATE SCHOOL CHOICE: THE EFFECTS OF RELIIOUS AFFILIATION AND PARTICIPATION Danny Cohen-Zada Department of Economcs, Ben-uron Unversty, Beer-Sheva 84105, Israel Wllam Sander Department of Economcs, DePaul

More information

To manage leave, meeting institutional requirements and treating individual staff members fairly and consistently.

To manage leave, meeting institutional requirements and treating individual staff members fairly and consistently. Corporate Polces & Procedures Human Resources - Document CPP216 Leave Management Frst Produced: Current Verson: Past Revsons: Revew Cycle: Apples From: 09/09/09 26/10/12 09/09/09 3 years Immedately Authorsaton:

More information

Calculating the high frequency transmission line parameters of power cables

Calculating the high frequency transmission line parameters of power cables < ' Calculatng the hgh frequency transmsson lne parameters of power cables Authors: Dr. John Dcknson, Laboratory Servces Manager, N 0 RW E B Communcatons Mr. Peter J. Ncholson, Project Assgnment Manager,

More information

BUSINESS PROCESS PERFORMANCE MANAGEMENT USING BAYESIAN BELIEF NETWORK. 0688, dskim@ssu.ac.kr

BUSINESS PROCESS PERFORMANCE MANAGEMENT USING BAYESIAN BELIEF NETWORK. 0688, dskim@ssu.ac.kr Proceedngs of the 41st Internatonal Conference on Computers & Industral Engneerng BUSINESS PROCESS PERFORMANCE MANAGEMENT USING BAYESIAN BELIEF NETWORK Yeong-bn Mn 1, Yongwoo Shn 2, Km Jeehong 1, Dongsoo

More information

The Application of Fractional Brownian Motion in Option Pricing

The Application of Fractional Brownian Motion in Option Pricing Vol. 0, No. (05), pp. 73-8 http://dx.do.org/0.457/jmue.05.0..6 The Applcaton of Fractonal Brownan Moton n Opton Prcng Qng-xn Zhou School of Basc Scence,arbn Unversty of Commerce,arbn zhouqngxn98@6.com

More information

THE DISTRIBUTION OF LOAN PORTFOLIO VALUE * Oldrich Alfons Vasicek

THE DISTRIBUTION OF LOAN PORTFOLIO VALUE * Oldrich Alfons Vasicek HE DISRIBUION OF LOAN PORFOLIO VALUE * Oldrch Alfons Vascek he amount of captal necessary to support a portfolo of debt securtes depends on the probablty dstrbuton of the portfolo loss. Consder a portfolo

More information

Dynamic Scheduling of Emergency Department Resources

Dynamic Scheduling of Emergency Department Resources Dynamc Schedulng of Emergency Department Resources Junchao Xao Laboratory for Internet Software Technologes, Insttute of Software, Chnese Academy of Scences P.O.Box 8718, No. 4 South Fourth Street, Zhong

More information

Latent Class Regression. Statistics for Psychosocial Research II: Structural Models December 4 and 6, 2006

Latent Class Regression. Statistics for Psychosocial Research II: Structural Models December 4 and 6, 2006 Latent Class Regresson Statstcs for Psychosocal Research II: Structural Models December 4 and 6, 2006 Latent Class Regresson (LCR) What s t and when do we use t? Recall the standard latent class model

More information

RELIABILITY, RISK AND AVAILABILITY ANLYSIS OF A CONTAINER GANTRY CRANE ABSTRACT

RELIABILITY, RISK AND AVAILABILITY ANLYSIS OF A CONTAINER GANTRY CRANE ABSTRACT Kolowrock Krzysztof Joanna oszynska MODELLING ENVIRONMENT AND INFRATRUCTURE INFLUENCE ON RELIABILITY AND OPERATION RT&A # () (Vol.) March RELIABILITY RIK AND AVAILABILITY ANLYI OF A CONTAINER GANTRY CRANE

More information

Fault tolerance in cloud technologies presented as a service

Fault tolerance in cloud technologies presented as a service Internatonal Scentfc Conference Computer Scence 2015 Pavel Dzhunev, PhD student Fault tolerance n cloud technologes presented as a servce INTRODUCTION Improvements n technques for vrtualzaton and performance

More information

Forecasting the Direction and Strength of Stock Market Movement

Forecasting the Direction and Strength of Stock Market Movement Forecastng the Drecton and Strength of Stock Market Movement Jngwe Chen Mng Chen Nan Ye cjngwe@stanford.edu mchen5@stanford.edu nanye@stanford.edu Abstract - Stock market s one of the most complcated systems

More information

Gender Classification for Real-Time Audience Analysis System

Gender Classification for Real-Time Audience Analysis System Gender Classfcaton for Real-Tme Audence Analyss System Vladmr Khryashchev, Lev Shmaglt, Andrey Shemyakov, Anton Lebedev Yaroslavl State Unversty Yaroslavl, Russa vhr@yandex.ru, shmaglt_lev@yahoo.com, andrey.shemakov@gmal.com,

More information

How To Know The Components Of Mean Squared Error Of Herarchcal Estmator S

How To Know The Components Of Mean Squared Error Of Herarchcal Estmator S S C H E D A E I N F O R M A T I C A E VOLUME 0 0 On Mean Squared Error of Herarchcal Estmator Stans law Brodowsk Faculty of Physcs, Astronomy, and Appled Computer Scence, Jagellonan Unversty, Reymonta

More information

Article received on April 23, 2007; accepted on October 18, 2007

Article received on April 23, 2007; accepted on October 18, 2007 A Renforcement Learnng Soluton for Allocatng Replcated Fragments n a Dstrbuted Database Una solucón de Aprendzae Reforzado para ubcar fragmentos replcados en Bases de Datos Dstrbudas Abel Rodríguez Morff

More information

A multiple objective test assembly approach for exposure control problems in Computerized Adaptive Testing

A multiple objective test assembly approach for exposure control problems in Computerized Adaptive Testing Pscológca (2), 3, 335-355. A multple objectve test assembly approach for exposure control problems n Computerzed Adaptve Testng Bernard P. Veldkamp * (), Angela J. Verschoor (2) & Theo J.H.M. Eggen (2)

More information