Introduction. 1. Mathematical formulation. 1.1 Standard formulation

Size: px
Start display at page:

Download "Introduction. 1. Mathematical formulation. 1.1 Standard formulation"

Transcription

1 Comact mathematcal formulaton for Grah Parttonng Marc BOULLE France Telecom R&D 2, Avenue Perre Marzn Lannon France Abstract. The grah arttonng roblem conssts of dvdng the vertces of a grah nto clusters, such that the weght of the edges crossng between clusters s mnmzed. We resent a new comact mathematcal formulaton of ths roblem, based on the use of bnary reresentaton for the ndex of clusters assgned to vertces. Ths new formulaton s almost mnmal n terms of the number of varables and constrants and of the densty of the constrant matrx. Its lnear relaxaton brngs a very fast comutatonal resoluton, comared wth the standard one. Exerments were conducted on classcal large benchmar grahs desgned for comarng heurstc methods. On one hand, these exerments show that the new formulaton s surrsngly less tme effcent than exected on general - arttonng roblems. On the other hand, the new formulaton aled on bsecton roblems allows to obtan the otmum soluton for about ten nstances, where only best uer bounds were revously nown. Key words. Lnear rogrammng, Grah Parttonng, Bsecton Introducton The grah arttonng s a classcal combnatoral otmzaton roblem. Ths toc has several alcatons n the telecommuncatons feld. For nstance, the frst ste n the desgn of telecommuncatons networ s to solve a arttonng roblem. The networ vertces are arttoned n order to mnmze the amount of traffc between clusters of vertces. Furthermore, the grah arttonng s an aealng otmzaton roblem, smle n ts defnton but NPcomlete (Garey, Johnson and Stocmeyer 1976). Parttonng a grah defned by ts vertces, ts edges and an edge cost functon conssts of dvdng the vertces nto clusters, such that the total weght of the edges whose endonts are n dfferent clusters s mnmzed. The number of clusters and ther sze are constraned. We consder the balanced artton of a grah, when the dfference of sze between clusters s at most one. The orgnalty of our formulaton s to use a bnary reresentaton for the ndex of clusters (assgned to vertces). The constrants n the standard arttonng formulaton must be deely reformulated to tae advantage of ths bnary reresentaton. The new formulaton results nto less number of varables, constrants and non null constrant coeffcents. It s very effcent for bsecton roblems and allowed to obtan the otmal soluton for several benchmar grahs not yet solved to otmalty. The remander of the document s organzed as follows: Secton one descrbes several new formulatons and demonstrates ther correctness. Secton two evaluates the comactness of the formulatons, and then the thrd secton resents exermental results obtaned by drectly mlementng the formulatons n a mxed nteger rogrammng solver and alyng them to classcal large benchmar grahs. 1. Mathematcal formulaton Let G=(V, E) be a grah wth vertex set V and edge set E. Let W j be the (ostve) weght of the edge (, j) between vertces and j, K the maxmum number of clusters and MaxCard the maxmum sze of each cluster ( V K. MaxCard ). The objectve s to mnmze the cut,.e. the total weght of the edges crossng the clusters. Ths s equvalent to maxmzng the total weght of the edges that are nsde the clusters. 1.1 Standard formulaton The standard straghtforward nteger rogrammng formulaton of the roblem s the followng.

2 Comact mathematcal formulaton for Grah Parttonng 2 Formulaton A Varables:. v =1 f vertex belongs to cluster and 0 otherwse. e j =1 f edge (, j) belongs to cluster and 0 otherwse Maxmze:, j, e j W j Subject to: Each vertex belongs to only one cluster (1) v = 1 Max sze of clusters (2) v MaxCard Lnearzaton of quadratc varables (e j =v *v j ) (3) e j v ej v j ej v + v j 1, j, Decson varables v bnary varables 1.2 Slght comact formulaton The recedng formulaton contans K varables and 3K constrants for each edge. Ths can be reduced to 1 varable and 2K constrants by ntroducng for each edge a new bnary varable f j valued 1 f the edge (,j) s nsde one cluster and 0 otherwse. We show below that varables f j can be evaluated from exstng vertex varables v. f e v v j = j = j f j = 1 / v = v j We can notce that v v v v = 1. Thus, f j can be exressed n the followng way: f j = Mn 1 j ( v v ) j j As the roblem of nterest s a maxmzaton roblem n f j wth ostve weghts, the recedng equalty constrant can be turned nto an nequalty constrant, and rewrtten n the followng way. fj Mn( 1 v v j ), fj 1 v v j Ths leads to the followng formulaton. Formulaton B Varables:. v =1 f vertex belongs to cluster and 0 otherwse. f j =1 f edge (, j) belongs to one of the clusters and 0 otherwse Maxmze:, j f j W j Subject to: Each vertex belongs to only one cluster (1) v = 1 Max sze of clusters (2) v MaxCard Constrants to calculate ntra-cluster edge varables (3) fj 1 + v v j f 1 v + v, j j j, Decson varables v bnary varables

3 Comact mathematcal formulaton for Grah Parttonng Alcaton to the bsecton roblem In the case of bsecton, one sngle decson varable s enough for each vertex. Its value s 1 f the vertex belongs to the frst cluster and 0 otherwse. Formulaton B can be smlfed n the followng way. Formulaton B2 Varables:. v 1 =1 f vertex belongs to frst cluster and 0 otherwse. f j =1 f edge (, j) belongs to one of the clusters and 0 otherwse Maxmze:, j f j W j Subject to: Max sze of clusters (1) MaxCard v 1 v 1 V MaxCard Constrants to calculate ntra-cluster edge varables (2) fj 1+ v1 v j1 fj 1 v 1 + v j1, j Decson varables v 1 bnary varables 1.4 Formulaton based on bnary ndexng of clusters In the general case of -arttonng, formulatons A and B need K decson varables for each vertex. These varables theoretcally enable 2 K choces for each vertex, although only K clusters are avalable. On ths bass, we wll am at a more comact formulaton, needng less varables and constrants. For each vertex, we have to decde to whch cluster t belongs. Let be an nteger varable corresondng to the ndex of the cluster assgned to vertex (0 <K). In ts bnary reresentaton, varable can be wrtten as a lnear combnaton of the owers of 2. The coeffcents b of ths lnear combnaton wll be used as decson varables. = b 2 (0 <log2(k)) For examle, f =5, =1*2 0 +0*2 1 +1*2 2, so b 0 =1=, b 1 =0, b 2 =1. Startng from formulaton B, we wll rewrte ts constrants wth these new decson varables. Each vertex belongs to only one cluster Ths constrant s satsfed from the defnton of, ndeendently of the choce of a decmal reresentaton (varables ) or a bnary reresentaton (varables b ). Each vertex must be assgned to an exstng cluster. Ths means that (ts bnary reresentaton) must be less than K-1. Max sze of clusters The sze of each cluster must be derved from the bnary decson varables corresondng to the cluster ndexes. Let us use agan varables v reresentng the assgnment of vertex to cluster. We wll show above how to calculate these varables from the new bnary varables b. The bnary reresentaton of ndex s based on constant values B : = B 2 Varable v equals 1 f and only f varable. equals. We then have the followng equvalences. v = 1 = v = 1, b = B b \ B v 1, 1 B 1 b + B b = ( )( ) 1 =

4 Comact mathematcal formulaton for Grah Parttonng 4 v ( 1 b B + 2b B ) 1 = 1 = So v s a roduct of factors a. a = 1 b B + 2bB (f B =0, a =1-b, otherwse a =b ) Varable v can be evaluated wth a generalzaton of the lnearzaton of quadratc varables., v a ( ) v a Thus, the avablty of varables v, derved from varables b, allows to reuse the same formula for the constrants concernng the max sze of clusters. Constrants to calculate ntra-cluster edge varables Intra cluster edge varables equal 1 f the endonts of the edge are n the same cluster,.e. f the ndexes of the clusters contanng the endonts of the edge have the same bnary reresentaton. f = 1 / b = b f j j = Mn 1 j ( b b ) j As the roblem of nterest s a maxmzaton roblem n f j wth ostve weghts, the recedng equalty constrant can be turned nto an nequalty constrant, and rewrtten n the followng way. f j ( 1 b b j ), fj b b j Mn 1 Formulaton C Let B be the coeffcent of 2 n the bnary reresentaton of ndex (B s a constant). Varables:. b : coeffcent of 2 n the bnary reresentaton of cluster ndex assgned to vertex. f j =1 f edge (, j) belongs to one of the clusters and 0 otherwse. v =1 f vertex belongs to cluster and 0 otherwse Maxmze: f j W j, j Subject to: Each vertex s assgned to a cluster ndex less than K (1) b 2 K 1 Max sze of clusters (2) v MaxCard Constrants to calculate ntra-cluster edge varables (3) fj 1 + b b j f 1 b + b, j j j, Lnearzaton of varables used for vertces assgnment to clusters (4) v 1 b B + 2b B, ( ) 1 + ( 1 b B + 2b B ) 1), v, Decson varables b bnary varables 1.5 Comact formulaton based on bnary ndexng of clusters Formulaton C uses few decson varables, but ncreases the number of constrants to calculate the sze of clusters. In order to decrease the number of varables and constrants used n formulaton C, we show n ths secton how to calculate the sze of clusters drectly from the varables b (bnary reresentaton of cluster ndexes), wthout usng revous varables v.

5 Comact mathematcal formulaton for Grah Parttonng 5 Frst, let us defne the noton of masng. An ndex s masng a bt f the bnary reresentaton of has bt set to 1. For examle, the ndexes 1 (1), 3 (11) and 5 (101) are masng bt 0. Let us now evaluate the term sb0 = b 0. sb 0 stands for the number of vertces assgned to a cluster whose ndex s masng bt 0. Thus, sb = b = cardcluster = cardcluster + cardcluster / s masng bt0 More generally, sb = b = s masng bt cardcluster 1 / The noton of masng can be generalzed to bt mass. An ndex s masng a bt mas m f the bnary reresentaton of has all bts of the bt mas m set to 1. For examle, wth m=5 (101), the ndexes 5 (101), 7 (111), 13 (1101) are masng the bt mas m. Let bm m be a bnary varable, equal to 1 f cluster ndex s masng bt mas m. Let sbm sbm =. = m bm m cardcluste m r / s masng bt mas m For examle, sbm 5 = cardcluster5 + cardcluster7 + cardcluster Ths leads to the followng rooston. Prooston 1: Varables sbm m ( 0 m < K ) can be lnearly derved from the szes of clusters cardcluster ( 0 < K ). Let SBM=M.C: sbm0 a00 a01... cardcluster0 sbm1 a10 a11 cardcluster1 = sbmk 1 ak 1.0 ak 1.1 cardclusterk 1 By defnton, coeffcent a m of matrx M equals 1 f the ndex s masng the bt mas m. Prooston 2: Matrx M s reversble. Proof: Each ndex s masng the bt mas, thus a =1. If s masng m, the bnary reresentaton of s above the bnary reresentaton of m for each bnary coeffcent, and thus m. Ths can be summarzed wth the followng roertes: a = 1 a m = 0 m < As a concluson, M s an uer dagonal matrx wth all ts dagonal terms set to 1. Hence, M s reversble. Corollary 1: The szes of clusters cardcluster ( 0 < K ) can be lnearly derved from the bt mas varables sbm m ( 0 m < K ). Prooston 2 demonstrates that the sze of clusters can be calculated from the new bt mas varables bm m. We show n the followng how to derve these bt mas varables bm m from the decson varables varables b. Frst, n the secal case of m=0, sbm0 = bm 0 = cardcluster = V. Second, we can notce that when the ndexes of clusters are owers of 2, there s a corresondence between the two sets of varables: bm m =b for m=2 ( 0 < log( K )). Thrd, for m > 0 and m 2, at least 2 bts n the bnary reresentaton of m are set to 1. Thus, m=m 1 +m 2, where m 1 s a ower of 2 (only one bt set to 1 n bnary reresentaton of m 1 ) and m 2 s strctly ostve. Based on the defnton of bt mas varables bm m, we get bm m = bm 1bm 2 and bm 1 corresonds to one of the decson varables b m m m. Therefore, the new bt mas varables bm m can be calculated by recurrence from the decson varables b. Usng the lnearzaton of quadratc varables, three constrants are necessary to calculate each new bt mas varable from two other varables (two decson varables, or one decson varable and one new smaller bt mas varable). 3

6 Comact mathematcal formulaton for Grah Parttonng 6 Let us llustrate these results wth an examle for K=6: sbm cardcluster0 sbm cardcluster1 sbm cardcluster2 = sbm cardcluster3 sbm cardcluster4 sbm cardcluster5 sbm 0 = V sbm 1, sbm 2, sbm 4 are calculated wth decson varables b 0, b 1, b 2 sbm 3 and sbm 5 are calculated wth bt mas varables bm 3 and bm 5 bm 3 =b 0.b 1 (3=11=01+10 n bnary reresentaton) bm 5 =b 0.b 2 (5=101= n bnary reresentaton) Formulaton D Let B be the coeffcent of 2 n the bnary reresentaton of ndex (B s a constant). Varables:. b : coeffcent of 2 n the bnary reresentaton of cluster ndex assgned to vertex. bm m =1 f the cluster ndex ndex s matchng the bt mas m. cardcluster : sze of cluster. f j =1 f edge (, j) belongs to one of the clusters and 0 otherwse Maxmze: f j W j, j Subject to: Each vertex s assgned to a cluster ndex less than K (1) b 2 K 1 Equatons to calculate the sze of clusters (2) cardcluster V b = = cardcluster / s masng bt bmm = / s masng bt mas m cardcluster m ( m 2 Max sze of clusters (3) cardcluster MaxCard Constrants to calculate ntra-cluster edge varables (4) fj 1 + b b j f 1 b + b, j j j, Lnearzaton of quadratc varables bm m (bm m =b bm m ) (5) bm m b bmm bm m' bm m b + bm m' 1 Decson varables b bnary varables, m ( m 2 2. Evaluaton of the sze of formulatons The szes of formulatons are evaluated n table 1 on the bass of the roblem dmensons: V : sze of V E : sze of E K: number of requred clusters. ) )

7 Comact mathematcal formulaton for Grah Parttonng 7 Formulaton Constrants Varables Bnary Non null coeffcents varables A (standard) V + 3 E K + V K + E K V K 2 V K + 7 E K B (A wth fewer constrants) K V + 2 E K + K B2 (B for K=2) 2 E + K C (bnary reresentaton) D (comact bnary reresentaton) Table 1. Sze of formulatons V (1+K+Klog(K)) + 2 E log(k) + K V (1+3(K-1-log(K))) + 2 E log(k) + 2K V K + E V + E V (K+log(K)) + E V (K-1) + E + K V K 2 V K + 6 E K V V K + 6 E V log(k) V (log(k)+2k+3klog(k)) + 6 E log(k) V log(k) V (2log(K) + 8(K-1-log(K))) + 6 E log(k) + K(K+1) Formulaton B s slghtly more comact than formulaton A. In the case of bsecton, formulaton B2 and D are almost dentcal. In ths case, usng only one bnary varable for cluster assgnment s actually the same as usng the bnary reresentaton of cluster ndexes. Thus, formulaton D can be seen as a generalzaton of formulaton B2 for greater K. Formulaton C s not nterestng. Ths s only an ntermedate ste to ntroduce formulaton D. Comarson of the sze of formulatons B and D The comactness of formulatons B and D vares n ooste way for V and E. E For the number of constrants: constr( D) constr( B) 1 + V 2 K log( K) log( K) 2V 1 K If E 3/2 V, the number of constrants n formulaton D s smaller than n formulaton B, and when E / V gets larger, the reducton factor n the number of constrants becomes close from log(k)/k. The number of varables s smlar n the two formulatons. However, the number of bnary varables s reduced wth a factor log(k)/k n formulaton D. E 4 1 K + 1 For the non null coeffcents: coef ( D) coef ( B) 1 + V 3 K log( K) log( K) 6V 1 K If E V, the number of non null coeffcents n formulaton D s smaller than the number of non null coeffcents n formulaton B, and when E / V gets larger, the reducton factor n the number of non null coeffcents becomes close from log(k)/k. As a concluson, formulaton D s more comact than formulaton B f E 3/2 V, and when E / V gets larger, the comactness reducton factor becomes close from log(k)/k for the number of constrants, of non null coeffcents and of bnary varables. Table 2 dslays a numercal examle for V =100, E =1000 and K=10. It shows a sgnfcant decrease n sze n the new formulaton for every dmenson of the roblem. Formulaton Constrants Varables Bnary varables Non null coeffcents A B D Table 2: Sze of formulatons for V =100, E =1000 and K=10 Remar: The value of the lnear relaxaton of formulatons A, B, B2, C and D s all edges of the grah. W j, j,.e. the sum of the weghts of

8 Comact mathematcal formulaton for Grah Parttonng 8 3. Resoluton wth a solver We comare the dfferent formulatons on the bass of numercal exerments by alyng a solver on a benchmar. 3.1 Benchmar nstances The nstances used for tests were ntroduced by (Johnson, Aragon, McGeoch and Schevon 1989). These nstances fall nto two categores: random grahs (resented n table 3) and random geometrc grahs (resented n table 4). In random grahs, edges are randomly generated n order to reach an average vertex degree. In random geometrc grahs, vertces are randomly located n square [0, 1]x[0, 1] and edges are created f ther length s less than a dstance d (d comuted n order to reach a gven average vertex degree). All weghts of edges are set to 1. V \ degree G G G G G G G G G G G G G G G G Table 3. Random grahs V \ degree U U U U U U U U Table 4. Random geometrc grahs 3.2 Resoluton envronment We used Clex 6.5 lbrary routnes to solve the mxed nteger formulaton drectly. For lnear relaxaton, we used the smlex dual method that roved much faster than the smlex rmal method for all formulatons. For exact resoluton (mxed nteger rogrammng), we actvated the roundng heurstc to set the ntal value of bnary varables, and the otons of memory swang of the branch and bound tree, not to be lmted by the hyscal memory of the machne. The target machne s a PC Dell Worstaton 400 wth Pentum II 300 MHz and 256 MB RAM, oeratng under Wndows/NT 4.0 system. Ths machne s rated 12.2 n SPECnt95 benchmar. 3.3 Slghtly stronger formulaton We can restrct the cluster ndexes of the vertces n the followng way: the frst vertex must belong to frst cluster the second vertex must belong ether to frst cluster, ether to second cluster the th vertex must belong to one of the frst clusters These constrants can be added to each formulaton: Each vertex wth ndex <K s assgned to a cluster wth ndex less than b 2 0 K 1 Wth these new constrants, the formulatons are less degenerated because ermutatons of vertex assgnment to clusters are now constraned. Exerments showed that the lnear relaxatons of formulatons were mroved wth only a few ercent comared wth the revously establshed relaxaton value. Exact resolutons were besdes twce faster thans to these new constrants. 3.4 Comarson of formulatons n lnear relaxaton resoluton The value of the bound obtaned wth lnear relaxaton s not reorted here. The only crteron used for comarson s CPU tme n seconds. We used random grah G that have 124 vertces and an average vertex degree 10 (620 edges) for ths comarson, whose results are dslayed n table 5.

9 Comact mathematcal formulaton for Grah Parttonng 9 Formulaton 3 clusters 4 clusters 5 clusters 6 clusters 8 clusters 16 clusters A B C D Table 5. CPU tme for -arttonng of nstance G wth lnear relaxaton Formulaton B s u to twce slower than formulaton A, although t s slghtly more comact. Formulaton D, based on bnary reresentaton of cluster ndexes, s much faster than standard formulaton A, wth a CPU tme reducton factor larger than ts comactness reducton factor log(k)/k 3.5 Comarson of formulatons n exact resoluton Exact resoluton s tractable wth all formulatons only wth small sze nstances. We frst evaluated all formulatons on two 2-arttonng and one 3-arttonng roblem. Results dslayed n table 6 show that standard formulaton A s clearly the slowest one. Formulaton D, whch s almost dentcal to formulaton B2, s the fastest on bsecton roblems. Surrsngly, n ste of ts comactness, formulaton D based on the bnary reresentaton of cluster ndexes s less effcent than formulaton B on the 3-arttonng roblem Formulaton Grah G clusters Grah G clusters Grah U clusters A B B C D Table 6. Exact resoluton CPU tme for 2 and 3-artonng of nstances G and U In order to confrm that trend, we roceeded wth other exerments of -arttonng and comared formulatons B and D on small random grahs wth 30 to 50 vertces, tractable wthn reasonable CPU tme. Table 7 reorts the number of bnary varables and the CPU tme for formulatons B and D, for each -arttonng roblem. In ste of ts comactness, formulaton D s always less tme effcent than formulaton B on general -arttonng roblems. Tests wth other solvers, usng dfferent resoluton strateges, would be necessary to confrm these results. Grah( V x E ) Clusters Formulaton B Formulaton D Bnary varables CPU tme Bnary varables CPU tme G1(24x26) G1(24x26) G1(24x26) G2(28x66) G2(28x66) G3(47x60) G3(47x60) Table7. Exact resoluton CPU tme for several -arttonng roblems on three small random grahs 3.6 Otmum soluton for large grahs Deste the dsaontng results for general -arttonng roblem, the resoluton tmes obtaned wth bsecton roblems were romsng. Ths was the reason for conductng new exerments to obtan as many as ossble otmum solutons on benchmar grahs. These grahs are large sze grahs whch are used to evaluate and to comare heurstc solvng methods. The urose of these last exerments s not to evaluate the dfferent formulatons, but to fnd otmal solutons on benchmar grahs. That s why we chose what seemed to be the fastest formulaton when used wth our straghtforward mlementaton n Clex solver. For bsecton roblems, we used formulaton B2, whch s almost dentcal to formulaton D n ths case. For -arttonng roblems, we chose formulaton B.

10 Comact mathematcal formulaton for Grah Parttonng 10 Bsecton The best nown values come from (Johnson, Aragon, McGeoch, Schevon 1989) for small nstances ( V 250) and from (Batttt and Bertoss 1999) for large nstances ( V 500). These uer bounds corresond to the best values obtaned by any heurstc method. Resolutons wth formulaton B2 were conducted on art of the nstances. Some of these resolutons were erformed untl otmalty was reached. Some other ones were stoed f otmalty was untractable. The last ones were not conducted at all when smaller smlar nstances were not solvable. The results are dslayed n the tables 8 and 9. The otmal cuts found wth exact resoluton are n bold face, wth the CPU tme (seconds). For the other nstances, the best nown uer bounds found by heurstc methods are reorted. V \ degree Cut CPU Cut CPU Cut CPU Cut CPU Table 8. Results for bsecton of random grahs wth formulaton B2 V \ degree Cut CPU Cut CPU Cut CPU Cut CPU Table 9. Results for bsecton of random geometrc grahs wth formulaton B2 Otmal cuts were obtaned on 9 out of the 24 nstances. These results obtaned wth formulaton B2 rove the otmalty of the corresondng uer bounds revously reached by heurstc methods. These new results (roven otmalty on 9 nstances) rovde an absolute crteron to evaluate heurstcs. It s surrsng that such large nstances could be solved to otmalty. The reason s that the solved nstances have a very low cut, and that ther lnear relaxaton s n ths case, very close to the otmum. Comutaton tme needed to rove otmalty ncreases qucly wth the sze of the grahs and the value of the cut. Random geometrc grahs are easer to solve than random grahs wth the nteger rogrammng formulaton. These grahs are more structured and thus the soluton sace s less degenerated than n the case of random grahs. Ths leads to a qucer exloraton of the branch and bound tree durng the resoluton. K-arttonng Otmal values were obtaned only wth two small random grahs, usng formulaton B. Results are reorted n tables 10 and 11 Grah G Otmal cut CPU tme 2 clusters clusters clusters clusters Table10. Results for -arttonng of nstance G wth formulaton B Grah G Otmal cut CPU tme 2 clusters clusters Table 11. Results for -arttonng of nstance G wth formulaton B These results rovde a benchmar for the -arttonng roblem. Comutaton tme ncreases extremely qucly wth the number of clusters. We used formulaton B because t was more effcent than formulaton D wth Clex solver. Formulaton D stll remans very romsng because of ts reduced sze (artcularly ts reduced number of bnary varables). We exect that usng formulaton D wth other solvers or wth a secally desgned resoluton method could mrove sgnfcantly the effcency of the resoluton, and therefore brng new otmum values.

11 Comact mathematcal formulaton for Grah Parttonng Qualty of the state of the art heurstcs The grah bsecton roblem has been extensvely studed n the ast, and many heurstcs have been exermented. For examle, algorthms such as (Kernghan-Ln 1970) or (Fducca-Mattheyes 1982) are frequently used to locally mrove bsectons. Many meta-heurstcs have also been used such as smulated annealng (Kratrc, Gellat and Vecch 1983) evaluated by (Johnson, Aragon, McGeoch and Schevon 1989), genetc algorthms used by (Bu and Moon 1996) or tabu search (Glover 1989) enhanced and adated to bsecton roblem by (Battt and Bertoss 1999). The multlevel aroach s secally ftted to very large grahs and constraned comutaton tme. It has been resented and studed by (Hendrcson and Leland 1995), (Monen and Demann 1997), (Pellegrn and Roman 1996), (Karys and Kumar 1998). These famles of heurstcs reresent a range of otons for the trade-off between comutaton tme and qualty of the soluton. In the lght of the 9 otmal solutons roven n ths aer, we shortly revew the results of some of the revously resented heurstcs, from the ont of vew of the qualty of solutons. The multlevel aroach that favors extremely short comutaton tme at the exense of the qualty of solutons s not evaluated. The frst evaluaton deals wth the results of (Johnson, Aragon, McGeoch and Schevon 1989), that ntroduced the benchmar nstances studed n ths aer and extensvely studed heurstcs based on smulated annealng. Table 12 comares otmal values wth best values found by the authors n 1989 wth ther smulated annealng based method. These early wors allowed to reach 6 of the 9 otmum cut szes. The 3 non otmal values are ndcated wth stars n table 12. Grah Otmal cut Best Smulated Annealng cut sze G G G G *52 U *4 U U U *3 U Table 12. Otmal versus best values for annealng method The second evaluaton s based on results resented by Bu and Moon 1996 for ther method based on genetc algorthms (BFS-GBA), and Battt and Bertoss 1999 for ther method based on tabu search (RRTS). The nstances wth less than 500 nodes are now consdered as too easy and are not taen nto account by the communty. Comutaton tmes have been scales wth resect to SecINT95 to be comarable. We comare the mnmal and average cuts of 1000 heurstc runs wth the otmal cut. Ths evaluaton shows that state of the art heurstcs have reached the otmal cut szes for these 6 nstances. The average cut sze of RRTS method s otmal for all nstances, excet for G Comutaton tme needed to rove the otmal soluton wth the exact resoluton method s about 1000 tmes that of heurstc methods. Grah Otmal resoluton BFS-GBA RRTS, 1000 n ter Otmal CPU Mn Average CPU Mn Average CPU G ,97 0,22 *51 52,06 0,98 U ,65 0, ,83 U ,68 0, ,32 U ,58 0, ,59 U ,78 0, ,05 U ,78 1, ,03 3,08 Table 13: Otmal versus best nown values for two state of the art methods based on genetc algorthms and tabu search Concluson The new formulaton of -arttonng roblem s more comact wth an average reducton factor of log(k)/k comared wth the standard formulaton. In artcular, the number of bnary varables used n the new formulaton s exactly reduced by a factor log(k)/k. Ths comact formulaton s romsng and could be the bass for effcent exact

12 Comact mathematcal formulaton for Grah Parttonng 12 resoluton methods or boundng methods. A straghtforward use of a commercal solver was conclusve for bsecton roblems, but resulted nto an ncreased comutaton tme for the resoluton of general -arttonng roblems. Ths new formulaton s nterestng, but further wor wll be necessary to conduct exerments wth other solvers or to desgn a dedcated exact resoluton method that could tae advantage of the comactness of the formulaton. Exerments were conducted on some large sze benchmar grahs, classcally used to comare heurstc methods. For all these nstances, revous wor reorted best nown values,.e. uer bounds. These exerments, based on the new formulaton for bsecton and on an almost standard formulaton for general -arttonng roblems, allowed to obtan more than 10 otmum values on nstances u to 1000 vertces and 5000 edges. These results rovde a strong evaluaton crtera to comare heurstc methods. References R. Battt and A.A. Bertoss, Greedy, Prohbton, and Reactve Heurstcs for Grah Parttonng, IEEE Transactons on Comuters, vol. 48, no T. N. Bu and B. R. Moon, Genetc Algorthm and Grah Parttonng, IEEE Transactons on Comuters, vol. 45, no C. Fducca and R. Mattheyses, A Lnear Tme Heurstc for Imrovng Networ Parttons, Proc. 19 th ACM/IEEE Desgn Automaton Conf., Las Vegas, M.R Garey, D.S. Johnson and L. Stocmeyer, Some smlfed NP-comlete grah roblems, Theoretcal Comuter Scence, 1, F. Glover, Tabu Search Part I, ORSA I Comutng, vol1, no3, B. Hendrcson and R. Leland, A multlevel algorthm for arttonng grahs, Proc. Suercomutng 95, ACM. D.S. Johnson, C.R. Aragon, L.A. McGeoch and C. Schevon, Otmzaton by Smulated Annealng: An Exermental Study; Part 1, Grah Parttonng, Oeratons Research, vol. 37, G. Karys and V. Kumar. (1998). A Fast and Hgh Qualty Multlevel Scheme for Parttonng Irregular Grahs, SIAM J. on Scentfc Comutng, to aear. B. Kernghan and S. Ln, An Effcent Heurstc Procedure for Parttonng Grahs, Bell Systems Techncal J., vol. 49, S. Kratrc, C.D. Gellat Jr. and M.P. Vecch, Otmzaton by Smulated Annealng, Scence, vol. 220, no. 4598, B. Monen and R. Demann, A Local Grah Parttonng Heurstc Meetng Bsecton Bounds, Proc. Eghth SIAM Conf. Parallel Processng for Scentfc Comutng. F. Pellegrn and J. Roman, Scotch: A Software Pacage for Statc Mang by Dual Recursve Barttonng of Process and Archtecture Grahs, Proc. HPCN 96 Brussels,

6.854 Advanced Algorithms Petar Maymounkov Problem Set 11 (November 23, 2005) With: Benjamin Rossman, Oren Weimann, and Pouya Kheradpour

6.854 Advanced Algorithms Petar Maymounkov Problem Set 11 (November 23, 2005) With: Benjamin Rossman, Oren Weimann, and Pouya Kheradpour 6.854 Advanced Algorthms Petar Maymounkov Problem Set 11 (November 23, 2005) Wth: Benjamn Rossman, Oren Wemann, and Pouya Kheradpour Problem 1. We reduce vertex cover to MAX-SAT wth weghts, such that the

More information

Support Vector Machines

Support Vector Machines Support Vector Machnes Decson surface s a hyperplane (lne n 2D) n feature space (smlar to the Perceptron) Arguably, the most mportant recent dscovery n machne learnng In a nutshell: map the data to a predetermned

More information

Parallelism for Nested Loops with Non-uniform and Flow Dependences

Parallelism for Nested Loops with Non-uniform and Flow Dependences Parallelsm for Nested Loops wth Non-unform and Flow Dependences Sam-Jn Jeong Dept. of Informaton & Communcaton Engneerng, Cheonan Unversty, 5, Anseo-dong, Cheonan, Chungnam, 330-80, Korea. seong@cheonan.ac.kr

More information

Support Vector Machines

Support Vector Machines /9/207 MIST.6060 Busness Intellgence and Data Mnng What are Support Vector Machnes? Support Vector Machnes Support Vector Machnes (SVMs) are supervsed learnng technques that analyze data and recognze patterns.

More information

On the Two-level Hybrid Clustering Algorithm

On the Two-level Hybrid Clustering Algorithm On the Two-level Clusterng Algorthm ng Yeow Cheu, Chee Keong Kwoh, Zongln Zhou Bonformatcs Research Centre, School of Comuter ngneerng, Nanyang Technologcal Unversty, Sngaore 639798 ezlzhou@ntu.edu.sg

More information

IMRT workflow. Optimization and Inverse planning. Intensity distribution IMRT IMRT. Dose optimization for IMRT. Bram van Asselen

IMRT workflow. Optimization and Inverse planning. Intensity distribution IMRT IMRT. Dose optimization for IMRT. Bram van Asselen IMRT workflow Otmzaton and Inverse lannng 69 Gy Bram van Asselen IMRT Intensty dstrbuton Webb 003: IMRT s the delvery of radaton to the atent va felds that have non-unform radaton fluence Purose: Fnd a

More information

Compiler Design. Spring Register Allocation. Sample Exercises and Solutions. Prof. Pedro C. Diniz

Compiler Design. Spring Register Allocation. Sample Exercises and Solutions. Prof. Pedro C. Diniz Compler Desgn Sprng 2014 Regster Allocaton Sample Exercses and Solutons Prof. Pedro C. Dnz USC / Informaton Scences Insttute 4676 Admralty Way, Sute 1001 Marna del Rey, Calforna 90292 pedro@s.edu Regster

More information

Lecture Note 08 EECS 4101/5101 Instructor: Andy Mirzaian. All Nearest Neighbors: The Lifting Method

Lecture Note 08 EECS 4101/5101 Instructor: Andy Mirzaian. All Nearest Neighbors: The Lifting Method Lecture Note 08 EECS 4101/5101 Instructor: Andy Mrzaan Introducton All Nearest Neghbors: The Lftng Method Suose we are gven aset P ={ 1, 2,..., n }of n onts n the lane. The gven coordnates of the -th ont

More information

Solving Optimization Problems on Orthogonal Ray Graphs

Solving Optimization Problems on Orthogonal Ray Graphs Solvng Otmzaton Problems on Orthogonal Ray Grahs Steven Chalck 1, Phl Kndermann 2, Faban L 2, Alexander Wolff 2 1 Insttut für Mathematk, TU Berln, Germany chalck@math.tu-berln.de 2 Lehrstuhl für Informatk

More information

Machine Learning. Support Vector Machines. (contains material adapted from talks by Constantin F. Aliferis & Ioannis Tsamardinos, and Martin Law)

Machine Learning. Support Vector Machines. (contains material adapted from talks by Constantin F. Aliferis & Ioannis Tsamardinos, and Martin Law) Machne Learnng Support Vector Machnes (contans materal adapted from talks by Constantn F. Alfers & Ioanns Tsamardnos, and Martn Law) Bryan Pardo, Machne Learnng: EECS 349 Fall 2014 Support Vector Machnes

More information

Course Introduction. Algorithm 8/31/2017. COSC 320 Advanced Data Structures and Algorithms. COSC 320 Advanced Data Structures and Algorithms

Course Introduction. Algorithm 8/31/2017. COSC 320 Advanced Data Structures and Algorithms. COSC 320 Advanced Data Structures and Algorithms Course Introducton Course Topcs Exams, abs, Proects A quc loo at a few algorthms 1 Advanced Data Structures and Algorthms Descrpton: We are gong to dscuss algorthm complexty analyss, algorthm desgn technques

More information

MODULE - 9 LECTURE NOTES 1 FUZZY OPTIMIZATION

MODULE - 9 LECTURE NOTES 1 FUZZY OPTIMIZATION Water Resources Systems Plannng an Management: vance Tocs Fuzzy Otmzaton MODULE - 9 LECTURE NOTES FUZZY OPTIMIZTION INTRODUCTION The moels scusse so far are crs an recse n nature. The term crs means chotonomous.e.,

More information

Optimized Query Planning of Continuous Aggregation Queries in Dynamic Data Dissemination Networks

Optimized Query Planning of Continuous Aggregation Queries in Dynamic Data Dissemination Networks WWW 007 / Trac: Performance and Scalablty Sesson: Scalable Systems for Dynamc Content Otmzed Query Plannng of Contnuous Aggregaton Queres n Dynamc Data Dssemnaton Networs Rajeev Guta IBM Inda Research

More information

Available online at ScienceDirect. Procedia Computer Science 94 (2016 )

Available online at  ScienceDirect. Procedia Computer Science 94 (2016 ) Avalable onlne at www.scencedrect.com ScenceDrect Proceda Comuter Scence 94 (2016 ) 176 182 The 13th Internatonal Conference on Moble Systems and Pervasve Comutng (MobSPC 2016) An Effcent QoS-aware Web

More information

Region Segmentation Readings: Chapter 10: 10.1 Additional Materials Provided

Region Segmentation Readings: Chapter 10: 10.1 Additional Materials Provided Regon Segmentaton Readngs: hater 10: 10.1 Addtonal Materals Provded K-means lusterng tet EM lusterng aer Grah Parttonng tet Mean-Shft lusterng aer 1 Image Segmentaton Image segmentaton s the oeraton of

More information

A New Approach For the Ranking of Fuzzy Sets With Different Heights

A New Approach For the Ranking of Fuzzy Sets With Different Heights New pproach For the ankng of Fuzzy Sets Wth Dfferent Heghts Pushpnder Sngh School of Mathematcs Computer pplcatons Thapar Unversty, Patala-7 00 Inda pushpndersnl@gmalcom STCT ankng of fuzzy sets plays

More information

Programming in Fortran 90 : 2017/2018

Programming in Fortran 90 : 2017/2018 Programmng n Fortran 90 : 2017/2018 Programmng n Fortran 90 : 2017/2018 Exercse 1 : Evaluaton of functon dependng on nput Wrte a program who evaluate the functon f (x,y) for any two user specfed values

More information

A new Algorithm for Lossless Compression applied to two-dimensional Static Images

A new Algorithm for Lossless Compression applied to two-dimensional Static Images A new Algorthm for Lossless Comresson aled to two-dmensonal Statc Images JUAN IGNACIO LARRAURI Deartment of Technology Industral Unversty of Deusto Avda. Unversdades, 4. 48007 Blbao SPAIN larrau@deusto.es

More information

Exercises (Part 4) Introduction to R UCLA/CCPR. John Fox, February 2005

Exercises (Part 4) Introduction to R UCLA/CCPR. John Fox, February 2005 Exercses (Part 4) Introducton to R UCLA/CCPR John Fox, February 2005 1. A challengng problem: Iterated weghted least squares (IWLS) s a standard method of fttng generalzed lnear models to data. As descrbed

More information

The Greedy Method. Outline and Reading. Change Money Problem. Greedy Algorithms. Applications of the Greedy Strategy. The Greedy Method Technique

The Greedy Method. Outline and Reading. Change Money Problem. Greedy Algorithms. Applications of the Greedy Strategy. The Greedy Method Technique //00 :0 AM Outlne and Readng The Greedy Method The Greedy Method Technque (secton.) Fractonal Knapsack Problem (secton..) Task Schedulng (secton..) Mnmum Spannng Trees (secton.) Change Money Problem Greedy

More information

Problem Set 3 Solutions

Problem Set 3 Solutions Introducton to Algorthms October 4, 2002 Massachusetts Insttute of Technology 6046J/18410J Professors Erk Demane and Shaf Goldwasser Handout 14 Problem Set 3 Solutons (Exercses were not to be turned n,

More information

A note on Schema Equivalence

A note on Schema Equivalence note on Schema Equvalence.H.M. ter Hofstede and H.. Proer and Th.P. van der Wede E.Proer@acm.org PUBLISHED S:.H.M. ter Hofstede, H.. Proer, and Th.P. van der Wede. Note on Schema Equvalence. Techncal Reort

More information

TPL-Aware Displacement-driven Detailed Placement Refinement with Coloring Constraints

TPL-Aware Displacement-driven Detailed Placement Refinement with Coloring Constraints TPL-ware Dsplacement-drven Detaled Placement Refnement wth Colorng Constrants Tao Ln Iowa State Unversty tln@astate.edu Chrs Chu Iowa State Unversty cnchu@astate.edu BSTRCT To mnmze the effect of process

More information

Application of Genetic Algorithms in Graph Theory and Optimization. Qiaoyan Yang, Qinghong Zeng

Application of Genetic Algorithms in Graph Theory and Optimization. Qiaoyan Yang, Qinghong Zeng 3rd Internatonal Conference on Materals Engneerng, Manufacturng Technology and Control (ICMEMTC 206) Alcaton of Genetc Algorthms n Grah Theory and Otmzaton Qaoyan Yang, Qnghong Zeng College of Mathematcs,

More information

Control strategies for network efficiency and resilience with route choice

Control strategies for network efficiency and resilience with route choice Control strateges for networ effcency and reslence wth route choce Andy Chow Ru Sha Centre for Transport Studes Unversty College London, UK Centralsed strateges UK 1 Centralsed strateges Some effectve

More information

THE CONDENSED FUZZY K-NEAREST NEIGHBOR RULE BASED ON SAMPLE FUZZY ENTROPY

THE CONDENSED FUZZY K-NEAREST NEIGHBOR RULE BASED ON SAMPLE FUZZY ENTROPY Proceedngs of the 20 Internatonal Conference on Machne Learnng and Cybernetcs, Guln, 0-3 July, 20 THE CONDENSED FUZZY K-NEAREST NEIGHBOR RULE BASED ON SAMPLE FUZZY ENTROPY JUN-HAI ZHAI, NA LI, MENG-YAO

More information

Parallel matrix-vector multiplication

Parallel matrix-vector multiplication Appendx A Parallel matrx-vector multplcaton The reduced transton matrx of the three-dmensonal cage model for gel electrophoress, descrbed n secton 3.2, becomes excessvely large for polymer lengths more

More information

5 The Primal-Dual Method

5 The Primal-Dual Method 5 The Prmal-Dual Method Orgnally desgned as a method for solvng lnear programs, where t reduces weghted optmzaton problems to smpler combnatoral ones, the prmal-dual method (PDM) has receved much attenton

More information

11. APPROXIMATION ALGORITHMS

11. APPROXIMATION ALGORITHMS Copng wth NP-completeness 11. APPROXIMATION ALGORITHMS load balancng center selecton prcng method: vertex cover LP roundng: vertex cover generalzed load balancng knapsack problem Q. Suppose I need to solve

More information

An Optimal Algorithm for Prufer Codes *

An Optimal Algorithm for Prufer Codes * J. Software Engneerng & Applcatons, 2009, 2: 111-115 do:10.4236/jsea.2009.22016 Publshed Onlne July 2009 (www.scrp.org/journal/jsea) An Optmal Algorthm for Prufer Codes * Xaodong Wang 1, 2, Le Wang 3,

More information

A Facet Generation Procedure. for solving 0/1 integer programs

A Facet Generation Procedure. for solving 0/1 integer programs A Facet Generaton Procedure for solvng 0/ nteger programs by Gyana R. Parja IBM Corporaton, Poughkeepse, NY 260 Radu Gaddov Emery Worldwde Arlnes, Vandala, Oho 45377 and Wlbert E. Wlhelm Teas A&M Unversty,

More information

Fuzzy Multi Objective Transportation Model. Based on New Ranking Index on. Generalized LR Fuzzy Numbers

Fuzzy Multi Objective Transportation Model. Based on New Ranking Index on. Generalized LR Fuzzy Numbers Aled Mathematcal Scences, Vol 8, 4, no 8, 6849-6879 HIKARI Ltd, wwwm-hkarcom htt://dxdoorg/988/ams4486 Fuzzy Mult Obectve Transortaton Model Based on New Rankng Index on Generalzed LR Fuzzy Numbers Y L

More information

GSLM Operations Research II Fall 13/14

GSLM Operations Research II Fall 13/14 GSLM 58 Operatons Research II Fall /4 6. Separable Programmng Consder a general NLP mn f(x) s.t. g j (x) b j j =. m. Defnton 6.. The NLP s a separable program f ts objectve functon and all constrants are

More information

On the Network Partitioning of Large Urban Transportation Networks

On the Network Partitioning of Large Urban Transportation Networks On the etwor Parttonng of Large Urban Transportaton etwors Hamdeh Etemadna and Khaled Abdelghany Abstract Ths paper ams at developng a traffc networ parttonng mechansm for dstrbuted traffc management applcatons.

More information

A Fast Content-Based Multimedia Retrieval Technique Using Compressed Data

A Fast Content-Based Multimedia Retrieval Technique Using Compressed Data A Fast Content-Based Multmeda Retreval Technque Usng Compressed Data Borko Furht and Pornvt Saksobhavvat NSF Multmeda Laboratory Florda Atlantc Unversty, Boca Raton, Florda 3343 ABSTRACT In ths paper,

More information

TIME-EFFICIENT NURBS CURVE EVALUATION ALGORITHMS

TIME-EFFICIENT NURBS CURVE EVALUATION ALGORITHMS TIME-EFFICIENT NURBS CURVE EVALUATION ALGORITHMS Kestuts Jankauskas Kaunas Unversty of Technology, Deartment of Multmeda Engneerng, Studentu st. 5, LT-5368 Kaunas, Lthuana, kestuts.jankauskas@ktu.lt Abstract:

More information

Classification / Regression Support Vector Machines

Classification / Regression Support Vector Machines Classfcaton / Regresson Support Vector Machnes Jeff Howbert Introducton to Machne Learnng Wnter 04 Topcs SVM classfers for lnearly separable classes SVM classfers for non-lnearly separable classes SVM

More information

U.C. Berkeley CS294: Beyond Worst-Case Analysis Handout 5 Luca Trevisan September 7, 2017

U.C. Berkeley CS294: Beyond Worst-Case Analysis Handout 5 Luca Trevisan September 7, 2017 U.C. Bereley CS294: Beyond Worst-Case Analyss Handout 5 Luca Trevsan September 7, 207 Scrbed by Haars Khan Last modfed 0/3/207 Lecture 5 In whch we study the SDP relaxaton of Max Cut n random graphs. Quc

More information

Broadcast Time Synchronization Algorithm for Wireless Sensor Networks Chaonong Xu 1)2)3), Lei Zhao 1)2), Yongjun Xu 1)2) and Xiaowei Li 1)2)

Broadcast Time Synchronization Algorithm for Wireless Sensor Networks Chaonong Xu 1)2)3), Lei Zhao 1)2), Yongjun Xu 1)2) and Xiaowei Li 1)2) Broadcast Tme Synchronzaton Algorthm for Wreless Sensor Networs Chaonong Xu )2)3), Le Zhao )2), Yongun Xu )2) and Xaowe L )2) ) Key Laboratory of Comuter Archtecture, Insttute of Comutng Technology Chnese

More information

Some material adapted from Mohamed Younis, UMBC CMSC 611 Spr 2003 course slides Some material adapted from Hennessy & Patterson / 2003 Elsevier

Some material adapted from Mohamed Younis, UMBC CMSC 611 Spr 2003 course slides Some material adapted from Hennessy & Patterson / 2003 Elsevier Some materal adapted from Mohamed Youns, UMBC CMSC 611 Spr 2003 course sldes Some materal adapted from Hennessy & Patterson / 2003 Elsever Scence Performance = 1 Executon tme Speedup = Performance (B)

More information

Machine Learning: Algorithms and Applications

Machine Learning: Algorithms and Applications 14/05/1 Machne Learnng: Algorthms and Applcatons Florano Zn Free Unversty of Bozen-Bolzano Faculty of Computer Scence Academc Year 011-01 Lecture 10: 14 May 01 Unsupervsed Learnng cont Sldes courtesy of

More information

An Application of the Dulmage-Mendelsohn Decomposition to Sparse Null Space Bases of Full Row Rank Matrices

An Application of the Dulmage-Mendelsohn Decomposition to Sparse Null Space Bases of Full Row Rank Matrices Internatonal Mathematcal Forum, Vol 7, 2012, no 52, 2549-2554 An Applcaton of the Dulmage-Mendelsohn Decomposton to Sparse Null Space Bases of Full Row Rank Matrces Mostafa Khorramzadeh Department of Mathematcal

More information

Hermite Splines in Lie Groups as Products of Geodesics

Hermite Splines in Lie Groups as Products of Geodesics Hermte Splnes n Le Groups as Products of Geodescs Ethan Eade Updated May 28, 2017 1 Introducton 1.1 Goal Ths document defnes a curve n the Le group G parametrzed by tme and by structural parameters n the

More information

An Efficient Algorithm for Minimum Vertex Cover Problem

An Efficient Algorithm for Minimum Vertex Cover Problem An Effcent Algorthm for Mnmum Vertex Cover Problem Rong Long Wang Zheng Tang Xn Shun Xu Member Non-member Non-member Paper An effcent parallel algorthm for solvng the mnmum vertex cover problem usng bnary

More information

MOBILE Cloud Computing (MCC) extends the capabilities

MOBILE Cloud Computing (MCC) extends the capabilities 1 Resource Sharng of a Computng Access Pont for Mult-user Moble Cloud Offloadng wth Delay Constrants Meng-Hs Chen, Student Member, IEEE, Mn Dong, Senor Member, IEEE, Ben Lang, Fellow, IEEE arxv:1712.00030v2

More information

Semi-Supervised Biased Maximum Margin Analysis for Interactive Image Retrieval

Semi-Supervised Biased Maximum Margin Analysis for Interactive Image Retrieval IP-06850-00.R3 Sem-Suervsed Based Maxmum Margn Analyss for Interactve Image Retreval Lnng Zhang,, Student Member, IEEE, Lo Wang, Senor Member, IEEE and Wes Ln 3, Senor Member, IEEE School of Electrcal

More information

Lecture 5: Multilayer Perceptrons

Lecture 5: Multilayer Perceptrons Lecture 5: Multlayer Perceptrons Roger Grosse 1 Introducton So far, we ve only talked about lnear models: lnear regresson and lnear bnary classfers. We noted that there are functons that can t be represented

More information

CS 534: Computer Vision Model Fitting

CS 534: Computer Vision Model Fitting CS 534: Computer Vson Model Fttng Sprng 004 Ahmed Elgammal Dept of Computer Scence CS 534 Model Fttng - 1 Outlnes Model fttng s mportant Least-squares fttng Maxmum lkelhood estmaton MAP estmaton Robust

More information

Support Vector Machines. CS534 - Machine Learning

Support Vector Machines. CS534 - Machine Learning Support Vector Machnes CS534 - Machne Learnng Perceptron Revsted: Lnear Separators Bnar classfcaton can be veed as the task of separatng classes n feature space: b > 0 b 0 b < 0 f() sgn( b) Lnear Separators

More information

LECTURE NOTES Duality Theory, Sensitivity Analysis, and Parametric Programming

LECTURE NOTES Duality Theory, Sensitivity Analysis, and Parametric Programming CEE 60 Davd Rosenberg p. LECTURE NOTES Dualty Theory, Senstvty Analyss, and Parametrc Programmng Learnng Objectves. Revew the prmal LP model formulaton 2. Formulate the Dual Problem of an LP problem (TUES)

More information

Uncertainty operations with Statool. Jianzhong Zhang. A thesis submitted to the graduate faculty

Uncertainty operations with Statool. Jianzhong Zhang. A thesis submitted to the graduate faculty Uncertanty oeratons wth Statool by Janzhong Zhang A thess submtted to the graduate faculty n artal fulfllment of the requrements for the degree of MASTER OF SCIECE Maor: Comutng Engneerng Program of Study

More information

Subspace clustering. Clustering. Fundamental to all clustering techniques is the choice of distance measure between data points;

Subspace clustering. Clustering. Fundamental to all clustering techniques is the choice of distance measure between data points; Subspace clusterng Clusterng Fundamental to all clusterng technques s the choce of dstance measure between data ponts; D q ( ) ( ) 2 x x = x x, j k = 1 k jk Squared Eucldean dstance Assumpton: All features

More information

SLAM Summer School 2006 Practical 2: SLAM using Monocular Vision

SLAM Summer School 2006 Practical 2: SLAM using Monocular Vision SLAM Summer School 2006 Practcal 2: SLAM usng Monocular Vson Javer Cvera, Unversty of Zaragoza Andrew J. Davson, Imperal College London J.M.M Montel, Unversty of Zaragoza. josemar@unzar.es, jcvera@unzar.es,

More information

Helsinki University Of Technology, Systems Analysis Laboratory Mat Independent research projects in applied mathematics (3 cr)

Helsinki University Of Technology, Systems Analysis Laboratory Mat Independent research projects in applied mathematics (3 cr) Helsnk Unversty Of Technology, Systems Analyss Laboratory Mat-2.08 Independent research projects n appled mathematcs (3 cr) "! #$&% Antt Laukkanen 506 R ajlaukka@cc.hut.f 2 Introducton...3 2 Multattrbute

More information

Sum of Linear and Fractional Multiobjective Programming Problem under Fuzzy Rules Constraints

Sum of Linear and Fractional Multiobjective Programming Problem under Fuzzy Rules Constraints Australan Journal of Basc and Appled Scences, 2(4): 1204-1208, 2008 ISSN 1991-8178 Sum of Lnear and Fractonal Multobjectve Programmng Problem under Fuzzy Rules Constrants 1 2 Sanjay Jan and Kalash Lachhwan

More information

A Multilevel Analytical Placement for 3D ICs

A Multilevel Analytical Placement for 3D ICs A Multlevel Analytcal Placement for 3D ICs Jason Cong, and Guoje Luo Computer Scence Department Unversty of Calforna, Los Angeles Calforna NanoSystems Insttute Los Angeles, CA 90095, USA Tel : (30) 06-775

More information

A NEW HEURISTIC APPROACH FOR DEMAND RESPONSIVE TRANSPORTATION SYSTEMS

A NEW HEURISTIC APPROACH FOR DEMAND RESPONSIVE TRANSPORTATION SYSTEMS A NEW HEURISTIC APPROACH FOR DEMAND RESPONSIVE TRANSPORTATION SYSTEMS Ru Jorge Gomes Faculdade de Engenhara da Unversdade do Porto Rua Dr. Roberto Fras, 4200-465 Porto, Portugal rjgomes@fe.u.t Jorge Pnho

More information

Overview. CSC 2400: Computer Systems. Pointers in C. Pointers - Variables that hold memory addresses - Using pointers to do call-by-reference in C

Overview. CSC 2400: Computer Systems. Pointers in C. Pointers - Variables that hold memory addresses - Using pointers to do call-by-reference in C CSC 2400: Comuter Systems Ponters n C Overvew Ponters - Varables that hold memory addresses - Usng onters to do call-by-reference n C Ponters vs. Arrays - Array names are constant onters Ponters and Strngs

More information

ACCURATE BIT ALLOCATION AND RATE CONTROL FOR DCT DOMAIN VIDEO TRANSCODING

ACCURATE BIT ALLOCATION AND RATE CONTROL FOR DCT DOMAIN VIDEO TRANSCODING ACCUATE BIT ALLOCATION AND ATE CONTOL FO DCT DOMAIN VIDEO TANSCODING Zhjun Le, Ncolas D. Georganas Multmeda Communcatons esearch Laboratory Unversty of Ottawa, Ottawa, Canada {lezj, georganas}@ mcrlab.uottawa.ca

More information

Edge Detection in Noisy Images Using the Support Vector Machines

Edge Detection in Noisy Images Using the Support Vector Machines Edge Detecton n Nosy Images Usng the Support Vector Machnes Hlaro Gómez-Moreno, Saturnno Maldonado-Bascón, Francsco López-Ferreras Sgnal Theory and Communcatons Department. Unversty of Alcalá Crta. Madrd-Barcelona

More information

Outline. Seamless Image Stitching in the Gradient Domain. Related Approaches. Image Stitching. Introduction Related Work

Outline. Seamless Image Stitching in the Gradient Domain. Related Approaches. Image Stitching. Introduction Related Work Outlne Seamless Image Sttchng n the Gradent Doman Anat Levn, Assaf Zomet, Shmuel Peleg and Yar Wess ECCV 004 Presenter: Pn Wu Oct 007 Introducton Related Work GIST: Gradent-doman Image Sttchng GIST GIST

More information

Harmonic Coordinates for Character Articulation PIXAR

Harmonic Coordinates for Character Articulation PIXAR Harmonc Coordnates for Character Artculaton PIXAR Pushkar Josh Mark Meyer Tony DeRose Bran Green Tom Sanock We have a complex source mesh nsde of a smpler cage mesh We want vertex deformatons appled to

More information

A Modelling and a New Hybrid MILP/CP Decomposition Method for Parallel Continuous Galvanizing Line Scheduling Problem

A Modelling and a New Hybrid MILP/CP Decomposition Method for Parallel Continuous Galvanizing Line Scheduling Problem ISIJ Internatonal, Vol. 58 (2018), ISIJ Internatonal, No. 10 Vol. 58 (2018), No. 10, pp. 1820 1827 A Modellng and a New Hybrd MILP/CP Decomposton Method for Parallel Contnuous Galvanzng Lne Schedulng Problem

More information

Minimization of the Expected Total Net Loss in a Stationary Multistate Flow Network System

Minimization of the Expected Total Net Loss in a Stationary Multistate Flow Network System Appled Mathematcs, 6, 7, 793-87 Publshed Onlne May 6 n ScRes. http://www.scrp.org/journal/am http://dx.do.org/.436/am.6.787 Mnmzaton of the Expected Total Net Loss n a Statonary Multstate Flow Networ System

More information

A MULTILEVEL DECISION MAKING MODEL FOR THE SUPPLIER SELECTION PROBLEM IN A FUZZY SITUATION

A MULTILEVEL DECISION MAKING MODEL FOR THE SUPPLIER SELECTION PROBLEM IN A FUZZY SITUATION OPERATIONS RESEARCH AND DECISIONS No. 4 2017 DOI: 10.5277/ord170401 Ahmad Yusuf ADHAMI 1 Syed Mohd MUNEEB 1 Mohammad Asm NOMANI 2 A MULTILEVEL DECISION MAKING MODEL FOR THE SUPPLIER SELECTION PROBLEM IN

More information

Problem Definitions and Evaluation Criteria for Computational Expensive Optimization

Problem Definitions and Evaluation Criteria for Computational Expensive Optimization Problem efntons and Evaluaton Crtera for Computatonal Expensve Optmzaton B. Lu 1, Q. Chen and Q. Zhang 3, J. J. Lang 4, P. N. Suganthan, B. Y. Qu 6 1 epartment of Computng, Glyndwr Unversty, UK Faclty

More information

An Iterative Solution Approach to Process Plant Layout using Mixed Integer Optimisation

An Iterative Solution Approach to Process Plant Layout using Mixed Integer Optimisation 17 th European Symposum on Computer Aded Process Engneerng ESCAPE17 V. Plesu and P.S. Agach (Edtors) 2007 Elsever B.V. All rghts reserved. 1 An Iteratve Soluton Approach to Process Plant Layout usng Mxed

More information

CMPS 10 Introduction to Computer Science Lecture Notes

CMPS 10 Introduction to Computer Science Lecture Notes CPS 0 Introducton to Computer Scence Lecture Notes Chapter : Algorthm Desgn How should we present algorthms? Natural languages lke Englsh, Spansh, or French whch are rch n nterpretaton and meanng are not

More information

Solving two-person zero-sum game by Matlab

Solving two-person zero-sum game by Matlab Appled Mechancs and Materals Onlne: 2011-02-02 ISSN: 1662-7482, Vols. 50-51, pp 262-265 do:10.4028/www.scentfc.net/amm.50-51.262 2011 Trans Tech Publcatons, Swtzerland Solvng two-person zero-sum game by

More information

Parallel Incremental Graph Partitioning Using Linear Programming

Parallel Incremental Graph Partitioning Using Linear Programming Syracuse Unversty SURFACE College of Engneerng and Computer Scence - Former Departments, Centers, Insttutes and roects College of Engneerng and Computer Scence 994 arallel Incremental Graph arttonng Usng

More information

Positive Semi-definite Programming Localization in Wireless Sensor Networks

Positive Semi-definite Programming Localization in Wireless Sensor Networks Postve Sem-defnte Programmng Localzaton n Wreless Sensor etworks Shengdong Xe 1,, Jn Wang, Aqun Hu 1, Yunl Gu, Jang Xu, 1 School of Informaton Scence and Engneerng, Southeast Unversty, 10096, anjng Computer

More information

Steps for Computing the Dissimilarity, Entropy, Herfindahl-Hirschman and. Accessibility (Gravity with Competition) Indices

Steps for Computing the Dissimilarity, Entropy, Herfindahl-Hirschman and. Accessibility (Gravity with Competition) Indices Steps for Computng the Dssmlarty, Entropy, Herfndahl-Hrschman and Accessblty (Gravty wth Competton) Indces I. Dssmlarty Index Measurement: The followng formula can be used to measure the evenness between

More information

MIXED INTEGER-DISCRETE-CONTINUOUS OPTIMIZATION BY DIFFERENTIAL EVOLUTION Part 1: the optimization method

MIXED INTEGER-DISCRETE-CONTINUOUS OPTIMIZATION BY DIFFERENTIAL EVOLUTION Part 1: the optimization method MIED INTEGER-DISCRETE-CONTINUOUS OPTIMIZATION BY DIFFERENTIAL EVOLUTION Part : the optmzaton method Joun Lampnen Unversty of Vaasa Department of Informaton Technology and Producton Economcs P. O. Box 700

More information

Optimization Methods: Integer Programming Integer Linear Programming 1. Module 7 Lecture Notes 1. Integer Linear Programming

Optimization Methods: Integer Programming Integer Linear Programming 1. Module 7 Lecture Notes 1. Integer Linear Programming Optzaton Methods: Integer Prograng Integer Lnear Prograng Module Lecture Notes Integer Lnear Prograng Introducton In all the prevous lectures n lnear prograng dscussed so far, the desgn varables consdered

More information

Hierarchical clustering for gene expression data analysis

Hierarchical clustering for gene expression data analysis Herarchcal clusterng for gene expresson data analyss Gorgo Valentn e-mal: valentn@ds.unm.t Clusterng of Mcroarray Data. Clusterng of gene expresson profles (rows) => dscovery of co-regulated and functonally

More information

Kent State University CS 4/ Design and Analysis of Algorithms. Dept. of Math & Computer Science LECT-16. Dynamic Programming

Kent State University CS 4/ Design and Analysis of Algorithms. Dept. of Math & Computer Science LECT-16. Dynamic Programming CS 4/560 Desgn and Analyss of Algorthms Kent State Unversty Dept. of Math & Computer Scence LECT-6 Dynamc Programmng 2 Dynamc Programmng Dynamc Programmng, lke the dvde-and-conquer method, solves problems

More information

A SYSTOLIC APPROACH TO LOOP PARTITIONING AND MAPPING INTO FIXED SIZE DISTRIBUTED MEMORY ARCHITECTURES

A SYSTOLIC APPROACH TO LOOP PARTITIONING AND MAPPING INTO FIXED SIZE DISTRIBUTED MEMORY ARCHITECTURES A SYSOLIC APPROACH O LOOP PARIIONING AND MAPPING INO FIXED SIZE DISRIBUED MEMORY ARCHIECURES Ioanns Drosts, Nektaros Kozrs, George Papakonstantnou and Panayots sanakas Natonal echncal Unversty of Athens

More information

Mathematics 256 a course in differential equations for engineering students

Mathematics 256 a course in differential equations for engineering students Mathematcs 56 a course n dfferental equatons for engneerng students Chapter 5. More effcent methods of numercal soluton Euler s method s qute neffcent. Because the error s essentally proportonal to the

More information

y and the total sum of

y and the total sum of Lnear regresson Testng for non-lnearty In analytcal chemstry, lnear regresson s commonly used n the constructon of calbraton functons requred for analytcal technques such as gas chromatography, atomc absorpton

More information

CS1100 Introduction to Programming

CS1100 Introduction to Programming Factoral (n) Recursve Program fact(n) = n*fact(n-) CS00 Introducton to Programmng Recurson and Sortng Madhu Mutyam Department of Computer Scence and Engneerng Indan Insttute of Technology Madras nt fact

More information

Maximum Feasibility Approach for Consensus Classifiers: Applications to Protein Structure Prediction

Maximum Feasibility Approach for Consensus Classifiers: Applications to Protein Structure Prediction Maxmum Feasblty Aroach for onsensus lassfers: Alcatons to Proten Structure Predcton Alesey Porollo 1, Rafal Adamcza 1, Mchael Wagner 1 and Jaroslaw Meller 1,2 1 Pedatrc Informatcs, 3333 Burnet Avenue,

More information

Obstacle-Aware Routing Problem in. a Rectangular Mesh Network

Obstacle-Aware Routing Problem in. a Rectangular Mesh Network Appled Mathematcal Scences, Vol. 9, 015, no. 14, 653-663 HIKARI Ltd, www.m-hkar.com http://dx.do.org/10.1988/ams.015.411911 Obstacle-Aware Routng Problem n a Rectangular Mesh Network Norazah Adzhar Department

More information

For instance, ; the five basic number-sets are increasingly more n A B & B A A = B (1)

For instance, ; the five basic number-sets are increasingly more n A B & B A A = B (1) Secton 1.2 Subsets and the Boolean operatons on sets If every element of the set A s an element of the set B, we say that A s a subset of B, or that A s contaned n B, or that B contans A, and we wrte A

More information

Data Representation in Digital Design, a Single Conversion Equation and a Formal Languages Approach

Data Representation in Digital Design, a Single Conversion Equation and a Formal Languages Approach Data Representaton n Dgtal Desgn, a Sngle Converson Equaton and a Formal Languages Approach Hassan Farhat Unversty of Nebraska at Omaha Abstract- In the study of data representaton n dgtal desgn and computer

More information

S.P.H. : A SOLUTION TO AVOID USING EROSION CRITERION?

S.P.H. : A SOLUTION TO AVOID USING EROSION CRITERION? S.P.H. : A SOLUTION TO AVOID USING EROSION CRITERION? Célne GALLET ENSICA 1 place Emle Bloun 31056 TOULOUSE CEDEX e-mal :cgallet@ensca.fr Jean Luc LACOME DYNALIS Immeuble AEROPOLE - Bat 1 5, Avenue Albert

More information

the nber of vertces n the graph. spannng tree T beng part of a par of maxmally dstant trees s called extremal. Extremal trees are useful n the mxed an

the nber of vertces n the graph. spannng tree T beng part of a par of maxmally dstant trees s called extremal. Extremal trees are useful n the mxed an On Central Spannng Trees of a Graph S. Bezrukov Unverstat-GH Paderborn FB Mathematk/Informatk Furstenallee 11 D{33102 Paderborn F. Kaderal, W. Poguntke FernUnverstat Hagen LG Kommunkatonssysteme Bergscher

More information

An Accurate Evaluation of Integrals in Convex and Non convex Polygonal Domain by Twelve Node Quadrilateral Finite Element Method

An Accurate Evaluation of Integrals in Convex and Non convex Polygonal Domain by Twelve Node Quadrilateral Finite Element Method Internatonal Journal of Computatonal and Appled Mathematcs. ISSN 89-4966 Volume, Number (07), pp. 33-4 Research Inda Publcatons http://www.rpublcaton.com An Accurate Evaluaton of Integrals n Convex and

More information

CSCI 104 Sorting Algorithms. Mark Redekopp David Kempe

CSCI 104 Sorting Algorithms. Mark Redekopp David Kempe CSCI 104 Sortng Algorthms Mark Redekopp Davd Kempe Algorthm Effcency SORTING 2 Sortng If we have an unordered lst, sequental search becomes our only choce If we wll perform a lot of searches t may be benefcal

More information

Smoothing Spline ANOVA for variable screening

Smoothing Spline ANOVA for variable screening Smoothng Splne ANOVA for varable screenng a useful tool for metamodels tranng and mult-objectve optmzaton L. Rcco, E. Rgon, A. Turco Outlne RSM Introducton Possble couplng Test case MOO MOO wth Game Theory

More information

Power-Aware Mapping for Network-on-Chip Architectures under Bandwidth and Latency Constraints

Power-Aware Mapping for Network-on-Chip Architectures under Bandwidth and Latency Constraints Power-Aware Mappng for Network-on-Chp Archtectures under Bandwdth and Latency Constrants Xaohang Wang 1,2, Me Yang 2, Yngtao Jang 2, and Peng Lu 1 1 Department of Informaton Scence and Electronc Engneerng,

More information

Energy and Throughput Optimized, Cluster Based Hierarchical Routing Algorithm for Heterogeneous Wireless Sensor Networks

Energy and Throughput Optimized, Cluster Based Hierarchical Routing Algorithm for Heterogeneous Wireless Sensor Networks Int. J. Communcatons Networ and System Scences 0 4 335-344 do:0.436/cns.0.45038 Publshed Onlne May 0 (htt://www.scrp.org/ournal/cns) Energy and Throughut Otmzed Cluster Based Herarchcal Routng Algorthm

More information

A Saturation Binary Neural Network for Crossbar Switching Problem

A Saturation Binary Neural Network for Crossbar Switching Problem A Saturaton Bnary Neural Network for Crossbar Swtchng Problem Cu Zhang 1, L-Qng Zhao 2, and Rong-Long Wang 2 1 Department of Autocontrol, Laonng Insttute of Scence and Technology, Benx, Chna bxlkyzhangcu@163.com

More information

QoS-Based Service Provision Schemes and Plan Durability in Service Composition

QoS-Based Service Provision Schemes and Plan Durability in Service Composition QoS-Based Servce Provson Schemes and Plan Durablty n Servce Comoston Koramt Pchanaharee and Twtte Senvongse Deartment of Comuter Engneerng, Faculty of Engneerng, Chulalongkorn Unversty Phyatha Road, Pathumwan,

More information

Image Segmentation. Image Segmentation

Image Segmentation. Image Segmentation Image Segmentaton REGION ORIENTED SEGMENTATION Let R reresent the entre mage regon. Segmentaton may be vewed as a rocess that arttons R nto n subregons, R, R,, Rn,such that n= R = R.e., the every xel must

More information

2x x l. Module 3: Element Properties Lecture 4: Lagrange and Serendipity Elements

2x x l. Module 3: Element Properties Lecture 4: Lagrange and Serendipity Elements Module 3: Element Propertes Lecture : Lagrange and Serendpty Elements 5 In last lecture note, the nterpolaton functons are derved on the bass of assumed polynomal from Pascal s trangle for the fled varable.

More information

Index Terms-Software effort estimation, principle component analysis, datasets, neural networks, and radial basis functions.

Index Terms-Software effort estimation, principle component analysis, datasets, neural networks, and radial basis functions. ISO 9001:2008 Certfed Internatonal Journal of Engneerng and Innovatve Technology (IJEIT The Effect of Dmensonalty Reducton on the Performance of Software Cost Estmaton Models Ryadh A.K. Mehd College of

More information

Physical synthesis for CPLD architectures

Physical synthesis for CPLD architectures Physcal synthess for CPLD archtectures Sd-Ahmed Senouc Mentor Grahcs, Grenoble, France Abstract In ths aer, we resent a new synthess feature namely, Xor matchng, and the foldback roduct term synthess for

More information

Unsupervised Learning and Clustering

Unsupervised Learning and Clustering Unsupervsed Learnng and Clusterng Why consder unlabeled samples?. Collectng and labelng large set of samples s costly Gettng recorded speech s free, labelng s tme consumng 2. Classfer could be desgned

More information

ON SOME ENTERTAINING APPLICATIONS OF THE CONCEPT OF SET IN COMPUTER SCIENCE COURSE

ON SOME ENTERTAINING APPLICATIONS OF THE CONCEPT OF SET IN COMPUTER SCIENCE COURSE Yordzhev K., Kostadnova H. Інформаційні технології в освіті ON SOME ENTERTAINING APPLICATIONS OF THE CONCEPT OF SET IN COMPUTER SCIENCE COURSE Yordzhev K., Kostadnova H. Some aspects of programmng educaton

More information

Needed Information to do Allocation

Needed Information to do Allocation Complexty n the Database Allocaton Desgn Must tae relatonshp between fragments nto account Cost of ntegrty enforcements Constrants on response-tme, storage, and processng capablty Needed Informaton to

More information