New l 1 -Norm Relaxations and Optimizations for Graph Clustering

Size: px
Start display at page:

Download "New l 1 -Norm Relaxations and Optimizations for Graph Clustering"

Transcription

1 Proceedngs of the Thrteth AAAI Conference on Artfcal Intellgence (AAAI-6) New l -Norm Relaxatons and Optmzatons for Graph Clusterng Fepng Ne, Hua Wang, Cheng Deng 3, Xnbo Gao 3, Xuelong L 4, Heng Huang Department of Computer Scence and Engneerng, Unversty of Texas at Arlngton, USA Department of Electrcal Engneerng and Computer Scence, Colorado School of Mnes, USA 3 School of Electronc Engneerng, Xdan Unversty, X an, Chna 4 X an Insttute of Optcs and Precson Mechancs, Chnese Academy of Scences, Chna fepngne@gmal.com, huawang@mnes.edu, {chdeng,xbgao}@mal.xdan.edu.cn, xuelong l@opt.ac.cn, heng@uta.edu Abstract In recent data mnng research, the graph clusterng methods, such as normalzed cut and rato cut, have been well studed and appled to solve many unsupervsed learnng applcatons. The orgnal graph clusterng methods are NP-hard problems. Tradtonal approaches used spectral relaxaton to solve the graph clusterng problems. The man dsadvantage of these approaches s that the obtaned spectral solutons could severely devate from the true soluton. To solve ths problem, n ths paper, we propose a new relaxaton mechansm for graph clusterng methods. Instead of mnmzng the squared dstances of clusterng results, we use the l -norm dstance. More mportant, consderng the normalzed consstency, we also use the l - norm for the normalzed terms n the new graph clusterng relaxatons. Due to the sparse result from the l -norm mnmzaton, the solutons of our new relaxed graph clusterng methods get dscrete values wth many zeros, whch are close to the deal solutons. Our new objectves are dffcult to be optmzed, because the mnmzaton problem nvolves the rato of nonsmooth terms. The exstng sparse learnng optmzaton algorthms cannot be appled to solve ths problem. In ths paper, we propose a new optmzaton algorthm to solve ths dffcult non-smooth rato mnmzaton problem. The extensve experments have been performed on three two-way clusterng and eght mult-way clusterng benchmark data sets. All emprcal results show that our new relaxaton methods consstently enhance the normalzed cut and rato cut clusterng results. Introducton Clusterng s an mportant task n computer vson and machne learnng research wth many applcatons, such as mage segmentaton (Sh and Malk ), mage categorzaton (Grauman and Darrell 6), scene analyss (Koppal and Narasmhan 6), moton modelng (P.Ochs and T.Brox ), and medcal mage analyss (Brun, Park, and To whom all correspondence should be addressed. Ths work was partally supported by the followng grants: NSF-IIS 7965, NSF-IIS 3675, NSF-IIS 3445, NSF-DBI 35668, NSF-IIS 4359, NIH R AG4937. Copyrght c 6, Assocaton for the Advancement of Artfcal Intellgence ( All rghts reserved. Shenton 4). In the past decades, many clusterng algorthms have been proposed. Among these approaches, the use of manfold nformaton n graph clusterng has shown the state-of-the-art clusterng performance. The graph based clusterng methods model the data as a weghted undrected graph based on the par-wse smlartes. Clusterng s then accomplshed by fndng the best cuts of the graph that optmze the predefned cost functons. Two types of graph clusterng methods, normalzed cut (Sh and Malk ) and rato cut (Cheng and We 99; Hagen and Kahng 99), are popularly used to solve the clusterng problems due to ther good clusterng performance. Solvng the graph clusterng problem s a dffcult task (NP-hard problems). The man dffculty of the graph clusterng problem comes from the constrans on the soluton. It s hard to solve the graph clusterng problems exactly. However, the approxmaton solutons are possble wth spectral relaxatons. The optmzaton usually leads to the computaton of the top egenvectors of certan graph affnty matrces, and the clusterng result can be derved from the obtaned egen-space. However, the tradtonal spectral relaxatons lead the non-optmal clusterng results. The spectral solutons don t drectly provde the clusterng results and the thresholdng post-processng has to be appled, such that the results often severely devate from the true soluton. More recently, tght relaxatons of balanced graph clusterng methods were proposed (Bühler and Hen 9; Luo et al. ; Hen and Setzer ), and gradent based method was used to solve the problem, whch s tme consumng and slow to converge n practce. In order to solve the above challengng ssues, n ths paper, we revst the normalzed cut and rato cut methods, and propose new relaxatons for these methods to acheve the dscrete and sparse clusterng results whch are close to the deal solutons. Instead of mnmzng the projected squared clusterng ndctors dstance, we mnmze the l dstance. Meanwhle, our new relaxatons also use the l -norm for the normalzaton terms. Due to the l -norm mnmzaton, most elements of each clusterng ndctor are enforced to be zero and hence the clusterng results are close the deal solutons. However, our new relatons ntroduce a dffcult optmzaton problem whch optmzes the rato of two nonsmooth terms. The standard optmzaton methods for sparse learnng, such as Proxmal Gradent, Iteratve Shrnkage- 96

2 Thresholdng, Gradent Projecton, Homotopy, and Augmented Lagrange Multpler methods, cannot be utlzed to solve such an l -norm rato mnmzaton problem. We propose a new optmzaton algorthm to solve ths dffcult problem wth theoretcally proved convergence, and our algorthm usually converges wthn teratons. The extensve clusterng experments are performed on three two-way clusterng data sets and eght mult-way clusterng data sets to evaluate our new relaxed normalzed cut and rato cut methods. All emprcal results demonstrate our new relaxatons consstently acheve better clusterng results than the tradtonal relaxatons. Graph Clusterng Revst Gven a graph G =(V,E) and the assocated weght matrx W, we partton t nto two dsjont sets A and B, A B = V, A B =, Two types of graph clusterng methods, normalzed cut (Sh and Malk ) and rato cut (Cheng and We 99; Hagen and Kahng 99), are usually appled to measure the qualty of the partton. The man task s to mnmze the defned graph cut to obtan a satsfed partton. Normalzed Cut and Relaxaton The normalzed cut (Sh and Malk ) s defned as cut(a, B) cut(a, B) Ncut = + assoc(a, V ) assoc(b,v ), () where cut(a, B) = A,j B W j and assoc(a, V ) = A,j V W j Denote a vector y R n as follows y =[,...,, r,...r] T. () }{{}}{{} n n Denote d = A D, d = B D, (Sh and Malk ) proved that when r = d d, the normalzed cut defned n Eq. () can be wrtten as Ncut = W j (y y j ),j D y = yt Ly y T Dy, (3) where L = D W s the Laplacan matrx, D s the dagonal matrx wth the -th dagonal element as D = j W j. Prevous paper (Sh and Malk ) provded proof, but here we provde a much more concse proof as follows. Let c = A,j B W j, then we have W j (y y j ),j D y = ( r) c d + r. (4) d On the other hand, accordng to Eq. (), we have Ncut = c + c. (5) d d Combnng the above equatons, we have: ( r) c d + r = c + c r + r d d d d + r = d + d d d d (d + rd ) = r = d d, whch completes the proof. In order to mnmze the normalzed cut to obtan a satsfed partton, we need to solve the followng problem: mn y=[,...,, d d,..., d d ] T W j (y y j ),j D y Due to the constrant on y, the problem s NP-hard. In order to solve ths problem, usually we need to relax the constrant. The constrant n Eq. (6) ndcates that T Dy =, thus the problem can be relaxed by usng the constrant T Dy = to replace the constrant n Eq. (6). The relaxed problem s as follows: mn T Dy= W j (y y j ),j D y The optmal soluton to the relaxed problem s the egenvector of D L correspondng to the second smallest egenvalue. However, ths relaxaton makes the soluton y devate from the constrant n Eq. (6) so much. The egenvector of D L usually take on contnuous values whle the real soluton of y should only take on two dscrete values. As suggested n (Sh and Malk ), One can take or the medan value as the splttng pont or one can search for the splttng pont such that the resultng partton has the best normalzed cut value. Rato Cut and Relaxaton The rato cut (Cheng and We 99; Hagen and Kahng 99) s defned as Rcut = cut(a, B) A + (6) (7) cut(a, B), (8) B where A denotes the number of ponts n A. Smlarly, t can be easly proved that when r = n n n Eq. (), the rato cut defned n Eq. (8) can be wrtten as W j (y y j ),j Rcut = = yt Ly y T y. (9) y In order to mnmze the normalzed cut to obtan a satsfed partton, we solve the followng problem W j (y y j ) mn y=[,...,, n n,..., n n ] T,j y () Due to the constrant on y, t was also proved that ths problem s NP-hard. The constrant n Eq. () ndcates that T y =, thus the problem can be relaxed by usng the constrant T y =to replace the constrant n Eq. (). The relaxed problem s as follows: mn T y= W j (y y j ),j y () 963

3 The optmal soluton to the relaxed problem s the egenvector of L correspondng to the second smallest egenvalue. The relaxaton also makes the soluton y devate from the constrant n Eq. (), and the fnal partton can be obtaned by the same strateges as n the case of normalzed cut. New Graph Clusterng Relaxatons and Optmzaton Algorthms As dscussed n the above secton, the tradtonal graph clusterng relaxatons make the soluton y devate from the deal soluton. In ths secton, we wll propose the new relaxatons for normalzed cut and rato cut, to whch the solutons are dscrete and close to the deal ones. We wll also provde new optmzaton algorthms to solve the proposed problems. New Relaxaton of Normalzed Cut Frst, we have the followng theorem for normalzed cut: Theorem Denote y = [,...,, d d,..., d d ] T, then W j y y j,j D y = Ncut Proof: As before, denote c = A,j B W j, then we have W j y y j d,j ( + d = )c = (d + d )c D y d d d = ( c + c )= d d Ncut, whch completes the proof. Based on Theorem, the problem (6) s equvalent to the followng problem wth the same constrant but dfferent objectve functon: W j y y j,j mn () y=[,...,, d d,..., d d ] D T y Accordngly, we can relax the problem as the followng one: W j y y j,j mn (3) T Dy= D y Note that problem (3) mnmzes a l -norm, whch usually results n sparse soluton (Ne et al. b). That s to say, y y j =for many (, j)-pars, whch ndcates the soluton y wll take on dscrete values. Therefore, the soluton to the relaxed problem (3) s close to the deal soluton. New Relaxaton of Rato Cut Smlarly, we have the followng theorem for rato cut: Theorem Denote y = [,...,, n n,..., n n ] T, then W j y y j,j y = Rcut Proof: As the above proof, denote c = A,j B W j, then we have: W j y y j n,j ( + n = )c = (n + n )c y n + n n n n n = ( c + c )= n n Rcut, whch completes the proof. Based on Theorem, the problem () s equvalent to the followng problem wth the same constrant but dfferent objectve functon: W j y y j,j mn (4) y=[,...,, n n,..., n n ] T y Accordngly, we can relax the problem as the followng one: W j y y j,j mn (5) T y= y Smlarly, the relaxed problem (5) wll result n sparse soluton,.e., y y j =for many (, j)-pars. Therefore, the soluton to the relaxed problem (5) s a good approxmaton to the deal soluton. Relaton to Cheeger cut In spectral graph theory (Chung 997), the Cheeger cut s defned as cut(a, B) Ccut = (6) mn{ A, B } As ponted by (Chung 997; Hen and Buhler ), the optmal Cheeger cut s the same as the value obtaned by optmal thresholdng the optmal soluton to the followng problem: mn y,medan(y)= W j y y j,j. (7) y Comparng Eq. (7) and Eq. (4), t s nterestng to see that the optmal Cheeger cut and the optmal rato cut can be obtaned wth the same objectve functon but under dfferent constrants. Note that the feasble soluton y to problem (7) can be contnuous values accordng to the constrant n Eq. (7), thus one can reasonably conjecture that the value obtaned by optmal thresholdng of the optmal soluton to problem (5) s close to the optmal rato cut n Eq. (4). Algorthms to Solve New Relaxaton Problems Our new relaxed graph clusterng methods ntroduce a dffcult optmzaton problem,.e. mnmze the rato of nonsmooth terms. The standard optmzaton methods for sparse learnng, such as Proxmal Gradent, Iteratve Shrnkage- Thresholdng, Gradent Projecton, Homotopy, and Augmented Lagrange Multpler methods, cannot be utlzed to 964

4 solve such l -norm rato mnmzaton problem. In ths secton, we wll propose a new optmzaton algorthm to solve ths challengng optmzaton problem. We frst ntroduce the soluton to a general problem, and then provde the solutons to problems n Eqs. (3) and (5), respectvely. A General Framework Before solvng the new relaxatons of graph clusterng methods, we solve the followng general problem frst: f (x) mn x C g (x). (8) Motvated by (Ne et al. 9; ; a; Ne, Yuan, and Huang 4), we gve an algorthm to solve ths problem, whch s very easy to mplement. The detaled algorthm s descrbed n Algorthm. In the followng, we wll prove that the algorthm wll monotoncally decrease the objectve value of problem (8) untl converges. Algorthm Algorthm to solve the general problem (8). Intalze x C whle not converge do. Calculate the objectve value λ = f (x) g (x). For each, calculate s = f and b (x) = sgn(g (x)). Update x by arg mn s f (x) λ b g (x) x C end whle Theorem 3 The procedure of Algorthm wll monotoncally decrease the objectve value of problem (8) untl converges. Proof: Denote the updated x by x. Accordng to step, s f ( x) λ b g ( x) s f (x) λ b g (x) Notce the defntons of s and b n step, we have f ( x) f (x) λ sgn(g (x))g ( x) f (x) λ g (x) It can be checked that the followng two nequaltes hold: ( f ( x) f ( x) ) f (x) (9) f (x) (sgn(g (x))g ( x) g ( x) ) () Addng the above three nequaltes n Eqs. (9-), we have f ( x) λ g ( x), whch ndcates f ( x) f (x) g ( x) λ = () g (x) Therefore, the algorthm wll monotoncally decrease the objectve value untl converges. Algorthm Algorthm to solve the problem (3). Intalze y such that T Dy = whle not converge do. Calculate λ = W j y y j,j D y ; the matrx S, where the (, j)-th element s S j = y y j ; and the vector b, where the -th element s b = sgn(d y ). Update y by y =arg mn T Dy= y T ˆLy λb T y, where ˆL = ˆD Ŵ, Ŵ = W S and ˆD s a dagonal matrx wth the -th element as ˆD = j Ŵj end whle Solutons to Problem (3) and Problem (5) We can use the algorthm framework n Algorthm to solve the proposed problem (3) and (5). The detaled algorthm to solve the problem (3) s descrbed n Algorthm. The algorthm to solve the problem (5) s smlar, we omt the detaled algorthm here durng to space lmtatons. In Step of the Algorthm, we need to solve the problem mn y T ˆLy λb T y. Solvng ths problem seems tme T Dy= consumng because of the constrant n the problem. Fortunately, the problem s equvalent to the followng problem mn y T ˆLy λb T y + ηy T D T Dy wth a large enough y η. Ths problem has a closed form soluton y = λ(ˆl + ηd T D) b and can be effcently solved by usng Woodbury matrx dentty and solvng a very sparse system of lnear equatons. Extenson to Mult-Way Parttonng The Algorthm parttons the graph nto two parts, we can recursvely run the algorthms to obtan the desred number of parttons. Specfcally, when the graph s dvded nto k parts, the k +part can be obtaned by runnng the algorthms on the k parts ndvdually, and select the one that the defned cut s mnmal when ths part s dvded nto parts. Another method to perform the mult-way parttonng s as follows. After we obtan k vectors by the algorthms, the k +vector y s obtaned by runnng the algorthms wth an addtonal constrant that the vector y s orthogonal to the pervous k vectors. Recursvely run the algorthms, we can obtan the desred number of vectors, and then run K-means clusterng on the vectors to obtan the fnal parttonng of the graph as n (Ne et al. b). Expermental Results In ths secton, we expermentally evaluate the two proposed graph clusterng methods n both two-way and mult-way clusterng tasks. We abbrevate the proposed new relaxaton of the normalzed cut as NR-NC, and abbrevate the proposed new relaxaton of the rato cut as NR-RC. To evaluate the clusterng results, we adopt the two wdely used standard metrcs: clusterng accuracy and normalzed mutual nformaton (NMI) (Ca et al. 8). 965

5 Table : Performance and objectve value comparson of the proposed methods aganst ther tradtonal counterparts. Rato Cut NR-RC Normalzed Cut NR-NC Data Acc NMI Acc NMI Acc NMI Acc NMI Hepatts onosphere breast cancer Logarthmc objectve value Rato Cut NR RC (our method) 3 4 Number of teratons (a) Objectve value vs. teraton. Average precson Number of teratons (b) Clusterng accuracy of NR-RC vs. teraton. Logarthmc objectve value Normalzed Cut NR NC (our method) 3 4 Number of teratons (c) Objectve value vs. teraton. Average precson Number of teratons (d) Clusterng accuracy NR-NC vs. teraton. Fgure : Convergence analyss of -way clusterng on hepatts data set. Logarthmc objectve value Rato Cut NR RC (our method) 3 4 Number of teratons (a) Objectve value vs. teraton. Average precson Number of teratons (b) Clusterng accuracy of NR-RC vs. teraton. Logarthmc objectve value Normalzed Cut NR NC (our method) 3 4 Number of teratons (c) Objectve value vs. teraton. Average precson Number of teratons (d) Clusterng accuracy NR-NC vs. teraton. Fgure : Convergence analyss of -way clusterng on onosphere data set. Logarthmc objectve value Rato Cut NR RC (our method) 3 4 Number of teratons (a) Objectve value vs. teraton. Average precson Number of teratons (b) Clusterng accuracy of NR-RC vs. teraton. Logarthmc objectve value Normalzed Cut NR NC (our method) 3 4 Number of teratons (c) Objectve value vs. teraton. Average precson Number of teratons (d) Clusterng accuracy NR-NC vs. teraton. Fgure 3: Convergence analyss of -way clusterng on breast cancer data set. Two-Way Clusterng Usng NR-RC and NR-NC Methods We frst evaluate the two proposed methods n two-way clusterng, and compare them aganst ther respectve tradtonal counterparts. Three benchmark data sets from UCI machne learnng repostory are used n our experments, ncludng hepatts database wth 55 nstances and attrbutes, onosphere database wth 35 nstances and 34 attrbutes, breast cancer database wth 86 nstances and 9 attrbutes. All these three data sets have only classes, therefore we can perform two-way clusterng on them. We construct nearestneghbor graph for each data set followng (Gu and Zhou 9). The clusterng results by the compared results are shown n Table, from whch we can see that the proposed new re- laxaton graph clusterng methods consstently outperforms ther tradtonal counterparts, sometmes very sgnfcantly. These results clearly demonstrate the advantage of the proposed methods n terms of clusterng performance. Because our methods employ teratve algorthms, we nvestgate the convergence propertes of our algorthms wth some detals. Gven the output vertex rankng from each teraton of the algorthms, we compute the objectve value by Eq. (8) for the NR-RC method and by Eq. () for the NR- NC method, whch are plotted n Fgure (a) and Fgure (c) for hepatts data, Fgure (a) and Fgure (c) for onosphere data, Fgure 3(a) and Fgure 3(c) for breast cancer data, respectvely. The clusterng accuracy wth respect each teraton of the two proposed methods are also plotted n Fgure (b) and Fgure (d) for hepatts data, Fgure (b) and Fgure (d) for onosphere data, Fgure 3(b) and Fgure 3(d) 966

6 Table : Clusterng accuracy (%) comparson of mult-way clusterng on the eght data sets. DATA SET KM PCA+KM LDA-KM RC NR-RC NC NR-NC DERMATOL ECOLI COIL BINALPHA UMIST AR YALEB PIE Table 3: NMI (%) comparson of mult-way clusterng on the eght data sets. DATA SET KM PCA+KM LDA-KM RC NR-RC NC NR-NC DERMATOL ECOLI COIL BINALPHA UMIST AR YALEB PIE for breast cancer data, respectvely. From these fgures we can see that our algorthms converge very fast wth typcally no more than teratons, whch concretely confrm ther computatonal effcency. Moreover, as shown n Fgure 3(a) and Fgure 3(c), n contrast to the objectve values of the tradtonal graph clusterng methods, the objectve values at convergence of our new relaxed graph clusterng methods are much smaller, whch provde another evdence to support the correctness of both our objectves and algorthms. Mult-Way Clusterng Usng NR-RC and NR-NC Methods Now we evaluate the proposed methods n mult-way clusterng. In our experments, we mplement our methods usng the second strategy ntroduced n Secton. Eght benchmark data sets are used n the experments, ncludng two UCI data sets, dermatology and ecol, one object data set, COIL- (Nene, Nayar, and Murase 996), one dgt and character data sets, Bnalpha, and four face data sets, Umst (Graham and Allnson 998), AR (Martnez and Benavente 998), YaleB (Georghades, Belhumeur, and Kregman ), and PIE (Sm and Baker 3). Besde comparng our methods to ther tradtonal counterparts, we also compare to K-means (denoted by Km), PCA+K-means (denoted by PCA+Km), LDA-Km (Dng and L 7) methods. Agan, we construct nearest-neghbor graph for each data set and set the neghborhood sze for graph constructon as (Gu and Zhou 9). The dmenson of PCA+K-means s searched from fve canddates rangng from to the dmenson of data. The results of all clusterng algorthms depend on the ntalzaton. To reduce statstcal varety, we ndependently repeat all clusterng algorthms for 5 tmes wth random ntalzatons, and then we report the results correspondng to the best objectve values. The clusterng performance measured by clusterng accuracy and NMI are reported n Table and Table 3, from whch we can see that the proposed methods stll perform the best among all compared methods. In addton, our methods are always better ther respectve tradtonal counterparts. These advantages valdate the effectveness of the proposed methods and justfy our motvatons. Conclusons In ths paper, we proposed new relaxatons for normalzed cut and rato cut methods. The l -norm dstances are utlzed n the relaxed graph clusterng formulatons. Such l -norm based relaxatons can naturally get the dscrete and sparse clusterng solutons (wth many zeros) whch are close to the optmal ones. Moreover, we proposed a new optmzaton algorthm to address the mnmzaton problem of a rato of non-smooth terms whch cannot be solved by other standard sparse learnng optmzaton algorthms. The valdatons were performed on both two-way and mult-way clusterng problems. On all eleven benchmark data sets, our new relaxed normalzed cut and rato cut methods consstently outperform the tradtonal ones. 967

7 References Brun, A.; Park, H.-J.; and Shenton, M. E. 4. Clusterng fber traces usng normalzed cuts. Medcal Image Computng and Computer-Asssted Interventon (MICCAI) Bühler, T., and Hen, M. 9. Spectral clusterng based on the graph p-laplacan. In Proceedngs of the 6th Annual Internatonal Conference on Machne Learnng, Ca, D.; He, X.; Wu, X.; and Han, J. 8. Non-negatve matrx factorzaton on manfold. In ICDM. Cheng, C.-K., and We, Y.-C. A. 99. An mproved twoway parttonng algorthm wth stable performance. IEEE Transactons on Computer-Aded Desgn of Integrated Crcuts and Systems ():5 5. Chung, F. R. K Spectral Graph Theory. CBMS Regonal Conference Seres n Mathematcs, No. 9, Amercan Mathematcal Socety. Dng, C. H. Q., and L, T. 7. Adaptve dmenson reducton usng dscrmnant analyss and -means clusterng. In ICML, Georghades, A.; Belhumeur, P.; and Kregman, D.. From few to many: Illumnaton cone models for face recognton under varable lghtng and pose. IEEE Transactons on PAMI 3(6): Graham, D. B., and Allnson, N. M Characterzng vrtual egensgnatures for general purpose face recognton. n face recognton: From theory to applcatons. NATO ASI Seres F, Computer and Systems Scences 63: Grauman, K., and Darrell, T. 6. Unsupervsed learnng of categores from sets of partally matchng mage features. In CVPR (), 9 5. Gu, Q., and Zhou, J. 9. Co-clusterng on manfolds. In SIGKDD. Hagen, L. W., and Kahng, A. B. 99. New spectral methods for rato cut parttonng and clusterng. IEEE Transactons on Computer-Aded Desgn of Integrated Crcuts and Systems (9): Hen, M., and Buhler, T.. An nverse power method for nonlnear egenproblems wth applcatons n -spectral clusterng and sparse PCA. In NIPS. Hen, M., and Setzer, S.. Beyond spectral clusterng - tght relaxatons of balanced graph cuts. In NIPS, Koppal, S., and Narasmhan, S. 6. Clusterng Appearance for Scene Analyss. CVPR. Luo, D.; Huang, H.; Dng, C. H. Q.; and Ne, F.. On the egenvectors of p-laplacan. Machne Learnng 8():37 5. Martnez, A. M., and Benavente, R The ar face database. In CVC Techncal Report. Nene, S. A.; Nayar, S. K.; and Murase, H Columba object mage lbrary (COIL-), Techncal Report CUCS Ne, F.; Xang, S.; Ja, Y.; and Zhang, C. 9. Semsupervsed orthogonal dscrmnant analyss va label propagaton. Pattern Recognton 4(): Ne, F.; Huang, H.; Ca, X.; and Dng, C.. Effcent and robust feature selecton va jont l, -norms mnmzaton. In NIPS. Ne, F.; Huang, H.; Dng, C.; Luo, D.; and Wang, H. a. Robust prncpal component analyss wth non-greedy l- norm maxmzaton. In IJCAI Proceedngs-Internatonal Jont Conference on Artfcal Intellgence, volume, 433. Ne, F.; Wang, H.; Huang, H.; and Dng, C. b. Unsupervsed and sem-supervsed learnng va l-norm graph. In IEEE Internatonal Conference on Computer Vson (ICCV), Ne, F.; Yuan, J.; and Huang, H. 4. Optmal mean robust prncpal component analyss. In Proceedngs of the 3st Internatonal Conference on Machne Learnng (ICML), 6 7. P.Ochs, and T.Brox.. Hgher order moton models and spectral clusterng. In IEEE Internatonal Conference on Computer Vson and Pattern Recognton (CVPR). Sh, J., and Malk, J.. Normalzed cuts and mage segmentaton. IEEE Transactons on PAMI (8): Sm, T., and Baker, S. 3. The cmu pose, llumnaton, and expresson database. IEEE Transactons on PAMI 5():

Learning the Kernel Parameters in Kernel Minimum Distance Classifier

Learning the Kernel Parameters in Kernel Minimum Distance Classifier Learnng the Kernel Parameters n Kernel Mnmum Dstance Classfer Daoqang Zhang 1,, Songcan Chen and Zh-Hua Zhou 1* 1 Natonal Laboratory for Novel Software Technology Nanjng Unversty, Nanjng 193, Chna Department

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

Cluster Analysis of Electrical Behavior

Cluster Analysis of Electrical Behavior Journal of Computer and Communcatons, 205, 3, 88-93 Publshed Onlne May 205 n ScRes. http://www.scrp.org/ournal/cc http://dx.do.org/0.4236/cc.205.350 Cluster Analyss of Electrcal Behavor Ln Lu Ln Lu, School

More information

The Research of Support Vector Machine in Agricultural Data Classification

The Research of Support Vector Machine in Agricultural Data Classification The Research of Support Vector Machne n Agrcultural Data Classfcaton Le Sh, Qguo Duan, Xnmng Ma, Me Weng College of Informaton and Management Scence, HeNan Agrcultural Unversty, Zhengzhou 45000 Chna Zhengzhou

More information

Classifier Selection Based on Data Complexity Measures *

Classifier Selection Based on Data Complexity Measures * Classfer Selecton Based on Data Complexty Measures * Edth Hernández-Reyes, J.A. Carrasco-Ochoa, and J.Fco. Martínez-Trndad Natonal Insttute for Astrophyscs, Optcs and Electroncs, Lus Enrque Erro No.1 Sta.

More information

Discriminative Dictionary Learning with Pairwise Constraints

Discriminative Dictionary Learning with Pairwise Constraints Dscrmnatve Dctonary Learnng wth Parwse Constrants Humn Guo Zhuoln Jang LARRY S. DAVIS UNIVERSITY OF MARYLAND Nov. 6 th, Outlne Introducton/motvaton Dctonary Learnng Dscrmnatve Dctonary Learnng wth Parwse

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

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

Determining the Optimal Bandwidth Based on Multi-criterion Fusion

Determining the Optimal Bandwidth Based on Multi-criterion Fusion Proceedngs of 01 4th Internatonal Conference on Machne Learnng and Computng IPCSIT vol. 5 (01) (01) IACSIT Press, Sngapore Determnng the Optmal Bandwdth Based on Mult-crteron Fuson Ha-L Lang 1+, Xan-Mn

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

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

Lecture 4: Principal components

Lecture 4: Principal components /3/6 Lecture 4: Prncpal components 3..6 Multvarate lnear regresson MLR s optmal for the estmaton data...but poor for handlng collnear data Covarance matrx s not nvertble (large condton number) Robustness

More information

Face Recognition University at Buffalo CSE666 Lecture Slides Resources:

Face Recognition University at Buffalo CSE666 Lecture Slides Resources: Face Recognton Unversty at Buffalo CSE666 Lecture Sldes Resources: http://www.face-rec.org/algorthms/ Overvew of face recognton algorthms Correlaton - Pxel based correspondence between two face mages Structural

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

LECTURE : MANIFOLD LEARNING

LECTURE : MANIFOLD LEARNING LECTURE : MANIFOLD LEARNING Rta Osadchy Some sldes are due to L.Saul, V. C. Raykar, N. Verma Topcs PCA MDS IsoMap LLE EgenMaps Done! Dmensonalty Reducton Data representaton Inputs are real-valued vectors

More information

Feature Reduction and Selection

Feature Reduction and Selection Feature Reducton and Selecton Dr. Shuang LIANG School of Software Engneerng TongJ Unversty Fall, 2012 Today s Topcs Introducton Problems of Dmensonalty Feature Reducton Statstc methods Prncpal Components

More information

Robust Dictionary Learning with Capped l 1 -Norm

Robust Dictionary Learning with Capped l 1 -Norm Proceedngs of the Twenty-Fourth Internatonal Jont Conference on Artfcal Intellgence (IJCAI 205) Robust Dctonary Learnng wth Capped l -Norm Wenhao Jang, Fepng Ne, Heng Huang Unversty of Texas at Arlngton

More information

Fast Robust Non-Negative Matrix Factorization for Large-Scale Human Action Data Clustering

Fast Robust Non-Negative Matrix Factorization for Large-Scale Human Action Data Clustering roceedngs of the Twenty-Ffth Internatonal Jont Conference on Artfcal Intellgence IJCAI-6) Fast Robust Non-Negatve Matrx Factorzaton for Large-Scale Human Acton Data Clusterng De Wang, Fepng Ne, Heng Huang

More information

A fast algorithm for color image segmentation

A fast algorithm for color image segmentation Unersty of Wollongong Research Onlne Faculty of Informatcs - Papers (Arche) Faculty of Engneerng and Informaton Scences 006 A fast algorthm for color mage segmentaton L. Dong Unersty of Wollongong, lju@uow.edu.au

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

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

Tsinghua University at TAC 2009: Summarizing Multi-documents by Information Distance

Tsinghua University at TAC 2009: Summarizing Multi-documents by Information Distance Tsnghua Unversty at TAC 2009: Summarzng Mult-documents by Informaton Dstance Chong Long, Mnle Huang, Xaoyan Zhu State Key Laboratory of Intellgent Technology and Systems, Tsnghua Natonal Laboratory for

More information

Content Based Image Retrieval Using 2-D Discrete Wavelet with Texture Feature with Different Classifiers

Content Based Image Retrieval Using 2-D Discrete Wavelet with Texture Feature with Different Classifiers IOSR Journal of Electroncs and Communcaton Engneerng (IOSR-JECE) e-issn: 78-834,p- ISSN: 78-8735.Volume 9, Issue, Ver. IV (Mar - Apr. 04), PP 0-07 Content Based Image Retreval Usng -D Dscrete Wavelet wth

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

Semi-Supervised Discriminant Analysis Based On Data Structure

Semi-Supervised Discriminant Analysis Based On Data Structure IOSR Journal of Computer Engneerng (IOSR-JCE) e-issn: 2278-0661,p-ISSN: 2278-8727, Volume 17, Issue 3, Ver. VII (May Jun. 2015), PP 39-46 www.osrournals.org Sem-Supervsed Dscrmnant Analyss Based On Data

More information

Outline. Type of Machine Learning. Examples of Application. Unsupervised Learning

Outline. Type of Machine Learning. Examples of Application. Unsupervised Learning Outlne Artfcal Intellgence and ts applcatons Lecture 8 Unsupervsed Learnng Professor Danel Yeung danyeung@eee.org Dr. Patrck Chan patrckchan@eee.org South Chna Unversty of Technology, Chna Introducton

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 Fast Visual Tracking Algorithm Based on Circle Pixels Matching

A Fast Visual Tracking Algorithm Based on Circle Pixels Matching A Fast Vsual Trackng Algorthm Based on Crcle Pxels Matchng Zhqang Hou hou_zhq@sohu.com Chongzhao Han czhan@mal.xjtu.edu.cn Ln Zheng Abstract: A fast vsual trackng algorthm based on crcle pxels matchng

More information

Human Face Recognition Using Generalized. Kernel Fisher Discriminant

Human Face Recognition Using Generalized. Kernel Fisher Discriminant Human Face Recognton Usng Generalzed Kernel Fsher Dscrmnant ng-yu Sun,2 De-Shuang Huang Ln Guo. Insttute of Intellgent Machnes, Chnese Academy of Scences, P.O.ox 30, Hefe, Anhu, Chna. 2. Department of

More information

Unsupervised Learning

Unsupervised Learning Pattern Recognton Lecture 8 Outlne Introducton Unsupervsed Learnng Parametrc VS Non-Parametrc Approach Mxture of Denstes Maxmum-Lkelhood Estmates Clusterng Prof. Danel Yeung School of Computer Scence and

More information

Recognizing Faces. Outline

Recognizing Faces. Outline Recognzng Faces Drk Colbry Outlne Introducton and Motvaton Defnng a feature vector Prncpal Component Analyss Lnear Dscrmnate Analyss !"" #$""% http://www.nfotech.oulu.f/annual/2004 + &'()*) '+)* 2 ! &

More information

MULTISPECTRAL IMAGES CLASSIFICATION BASED ON KLT AND ATR AUTOMATIC TARGET RECOGNITION

MULTISPECTRAL IMAGES CLASSIFICATION BASED ON KLT AND ATR AUTOMATIC TARGET RECOGNITION MULTISPECTRAL IMAGES CLASSIFICATION BASED ON KLT AND ATR AUTOMATIC TARGET RECOGNITION Paulo Quntlano 1 & Antono Santa-Rosa 1 Federal Polce Department, Brasla, Brazl. E-mals: quntlano.pqs@dpf.gov.br and

More information

Collaboratively Regularized Nearest Points for Set Based Recognition

Collaboratively Regularized Nearest Points for Set Based Recognition Academc Center for Computng and Meda Studes, Kyoto Unversty Collaboratvely Regularzed Nearest Ponts for Set Based Recognton Yang Wu, Mchhko Mnoh, Masayuk Mukunok Kyoto Unversty 9/1/013 BMVC 013 @ Brstol,

More information

Nonlocal Mumford-Shah Model for Image Segmentation

Nonlocal Mumford-Shah Model for Image Segmentation for Image Segmentaton 1 College of Informaton Engneerng, Qngdao Unversty, Qngdao, 266000,Chna E-mal:ccluxaoq@163.com ebo e 23 College of Informaton Engneerng, Qngdao Unversty, Qngdao, 266000,Chna E-mal:

More information

Graph-based Clustering

Graph-based Clustering Graphbased Clusterng Transform the data nto a graph representaton ertces are the data ponts to be clustered Edges are eghted based on smlarty beteen data ponts Graph parttonng Þ Each connected component

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

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

A Binarization Algorithm specialized on Document Images and Photos

A Binarization Algorithm specialized on Document Images and Photos A Bnarzaton Algorthm specalzed on Document mages and Photos Ergna Kavalleratou Dept. of nformaton and Communcaton Systems Engneerng Unversty of the Aegean kavalleratou@aegean.gr Abstract n ths paper, a

More information

Tone-Aware Sparse Representation for Face Recognition

Tone-Aware Sparse Representation for Face Recognition Tone-Aware Sparse Representaton for Face Recognton Lngfeng Wang, Huayu Wu and Chunhong Pan Abstract It s stll a very challengng task to recognze a face n a real world scenaro, snce the face may be corrupted

More information

Unsupervised Learning and Clustering

Unsupervised Learning and Clustering Unsupervsed Learnng and Clusterng Supervsed vs. Unsupervsed Learnng Up to now we consdered supervsed learnng scenaro, where we are gven 1. samples 1,, n 2. class labels for all samples 1,, n Ths s also

More information

Face Recognition Based on SVM and 2DPCA

Face Recognition Based on SVM and 2DPCA Vol. 4, o. 3, September, 2011 Face Recognton Based on SVM and 2DPCA Tha Hoang Le, Len Bu Faculty of Informaton Technology, HCMC Unversty of Scence Faculty of Informaton Scences and Engneerng, Unversty

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

CHAPTER 3 SEQUENTIAL MINIMAL OPTIMIZATION TRAINED SUPPORT VECTOR CLASSIFIER FOR CANCER PREDICTION

CHAPTER 3 SEQUENTIAL MINIMAL OPTIMIZATION TRAINED SUPPORT VECTOR CLASSIFIER FOR CANCER PREDICTION 48 CHAPTER 3 SEQUENTIAL MINIMAL OPTIMIZATION TRAINED SUPPORT VECTOR CLASSIFIER FOR CANCER PREDICTION 3.1 INTRODUCTION The raw mcroarray data s bascally an mage wth dfferent colors ndcatng hybrdzaton (Xue

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

Improvement of Spatial Resolution Using BlockMatching Based Motion Estimation and Frame. Integration

Improvement of Spatial Resolution Using BlockMatching Based Motion Estimation and Frame. Integration Improvement of Spatal Resoluton Usng BlockMatchng Based Moton Estmaton and Frame Integraton Danya Suga and Takayuk Hamamoto Graduate School of Engneerng, Tokyo Unversty of Scence, 6-3-1, Nuku, Katsuska-ku,

More information

Network Intrusion Detection Based on PSO-SVM

Network Intrusion Detection Based on PSO-SVM TELKOMNIKA Indonesan Journal of Electrcal Engneerng Vol.1, No., February 014, pp. 150 ~ 1508 DOI: http://dx.do.org/10.11591/telkomnka.v1.386 150 Network Intruson Detecton Based on PSO-SVM Changsheng Xang*

More information

Simplification of 3D Meshes

Simplification of 3D Meshes Smplfcaton of 3D Meshes Addy Ngan /4/00 Outlne Motvaton Taxonomy of smplfcaton methods Hoppe et al, Mesh optmzaton Hoppe, Progressve meshes Smplfcaton of 3D Meshes 1 Motvaton Hgh detaled meshes becomng

More information

BOOSTING CLASSIFICATION ACCURACY WITH SAMPLES CHOSEN FROM A VALIDATION SET

BOOSTING CLASSIFICATION ACCURACY WITH SAMPLES CHOSEN FROM A VALIDATION SET 1 BOOSTING CLASSIFICATION ACCURACY WITH SAMPLES CHOSEN FROM A VALIDATION SET TZU-CHENG CHUANG School of Electrcal and Computer Engneerng, Purdue Unversty, West Lafayette, Indana 47907 SAUL B. GELFAND School

More information

High-Boost Mesh Filtering for 3-D Shape Enhancement

High-Boost Mesh Filtering for 3-D Shape Enhancement Hgh-Boost Mesh Flterng for 3-D Shape Enhancement Hrokazu Yagou Λ Alexander Belyaev y Damng We z Λ y z ; ; Shape Modelng Laboratory, Unversty of Azu, Azu-Wakamatsu 965-8580 Japan y Computer Graphcs Group,

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

Face Recognition by Fusing Binary Edge Feature and Second-order Mutual Information

Face Recognition by Fusing Binary Edge Feature and Second-order Mutual Information Face Recognton by Fusng Bnary Edge Feature and Second-order Mutual Informaton Jatao Song, Bejng Chen, We Wang, Xaobo Ren School of Electronc and Informaton Engneerng, Nngbo Unversty of Technology Nngbo,

More information

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

Local Quaternary Patterns and Feature Local Quaternary Patterns

Local Quaternary Patterns and Feature Local Quaternary Patterns Local Quaternary Patterns and Feature Local Quaternary Patterns Jayu Gu and Chengjun Lu The Department of Computer Scence, New Jersey Insttute of Technology, Newark, NJ 0102, USA Abstract - Ths paper presents

More information

Classifying Acoustic Transient Signals Using Artificial Intelligence

Classifying Acoustic Transient Signals Using Artificial Intelligence Classfyng Acoustc Transent Sgnals Usng Artfcal Intellgence Steve Sutton, Unversty of North Carolna At Wlmngton (suttons@charter.net) Greg Huff, Unversty of North Carolna At Wlmngton (jgh7476@uncwl.edu)

More information

Fitting: Deformable contours April 26 th, 2018

Fitting: Deformable contours April 26 th, 2018 4/6/08 Fttng: Deformable contours Aprl 6 th, 08 Yong Jae Lee UC Davs Recap so far: Groupng and Fttng Goal: move from array of pxel values (or flter outputs) to a collecton of regons, objects, and shapes.

More information

Laplacian Eigenmap for Image Retrieval

Laplacian Eigenmap for Image Retrieval Laplacan Egenmap for Image Retreval Xaofe He Partha Nyog Department of Computer Scence The Unversty of Chcago, 1100 E 58 th Street, Chcago, IL 60637 ABSTRACT Dmensonalty reducton has been receved much

More information

A B-Snake Model Using Statistical and Geometric Information - Applications to Medical Images

A B-Snake Model Using Statistical and Geometric Information - Applications to Medical Images A B-Snake Model Usng Statstcal and Geometrc Informaton - Applcatons to Medcal Images Yue Wang, Eam Khwang Teoh and Dnggang Shen 2 School of Electrcal and Electronc Engneerng, Nanyang Technologcal Unversty

More information

An Improved Spectral Clustering Algorithm Based on Local Neighbors in Kernel Space 1

An Improved Spectral Clustering Algorithm Based on Local Neighbors in Kernel Space 1 DOI: 10.98/CSIS110415064L An Improved Spectral Clusterng Algorthm Based on Local Neghbors n Kernel Space 1 Xnyue Lu 1,, Xng Yong and Hongfe Ln 1 1 School of Computer Scence and Technology, Dalan Unversty

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

Small Network Segmentation with Template Guidance

Small Network Segmentation with Template Guidance Small Network Segmentaton wth Template Gudance Krstn Dane Lu Department of Mathematcs Unversty of Calforna, Davs Davs, CA 95616 kdlu@math.ucdavs.edu Ian Davdson Department of Computer Scence Unversty of

More information

Term Weighting Classification System Using the Chi-square Statistic for the Classification Subtask at NTCIR-6 Patent Retrieval Task

Term Weighting Classification System Using the Chi-square Statistic for the Classification Subtask at NTCIR-6 Patent Retrieval Task Proceedngs of NTCIR-6 Workshop Meetng, May 15-18, 2007, Tokyo, Japan Term Weghtng Classfcaton System Usng the Ch-square Statstc for the Classfcaton Subtask at NTCIR-6 Patent Retreval Task Kotaro Hashmoto

More information

S1 Note. Basis functions.

S1 Note. Basis functions. S1 Note. Bass functons. Contents Types of bass functons...1 The Fourer bass...2 B-splne bass...3 Power and type I error rates wth dfferent numbers of bass functons...4 Table S1. Smulaton results of type

More information

Competitive Sparse Representation Classification for Face Recognition

Competitive Sparse Representation Classification for Face Recognition Vol. 6, No. 8, 05 Compettve Sparse Representaton Classfcaton for Face Recognton Yng Lu Chongqng Key Laboratory of Computatonal Intellgence Chongqng Unversty of Posts and elecommuncatons Chongqng, Chna

More information

BAYESIAN MULTI-SOURCE DOMAIN ADAPTATION

BAYESIAN MULTI-SOURCE DOMAIN ADAPTATION BAYESIAN MULTI-SOURCE DOMAIN ADAPTATION SHI-LIANG SUN, HONG-LEI SHI Department of Computer Scence and Technology, East Chna Normal Unversty 500 Dongchuan Road, Shangha 200241, P. R. Chna E-MAIL: slsun@cs.ecnu.edu.cn,

More information

A Novel Adaptive Descriptor Algorithm for Ternary Pattern Textures

A Novel Adaptive Descriptor Algorithm for Ternary Pattern Textures A Novel Adaptve Descrptor Algorthm for Ternary Pattern Textures Fahuan Hu 1,2, Guopng Lu 1 *, Zengwen Dong 1 1.School of Mechancal & Electrcal Engneerng, Nanchang Unversty, Nanchang, 330031, Chna; 2. School

More information

NUMERICAL SOLVING OPTIMAL CONTROL PROBLEMS BY THE METHOD OF VARIATIONS

NUMERICAL SOLVING OPTIMAL CONTROL PROBLEMS BY THE METHOD OF VARIATIONS ARPN Journal of Engneerng and Appled Scences 006-017 Asan Research Publshng Network (ARPN). All rghts reserved. NUMERICAL SOLVING OPTIMAL CONTROL PROBLEMS BY THE METHOD OF VARIATIONS Igor Grgoryev, Svetlana

More information

From Comparing Clusterings to Combining Clusterings

From Comparing Clusterings to Combining Clusterings Proceedngs of the Twenty-Thrd AAAI Conference on Artfcal Intellgence (008 From Comparng Clusterngs to Combnng Clusterngs Zhwu Lu and Yuxn Peng and Janguo Xao Insttute of Computer Scence and Technology,

More information

Active Contours/Snakes

Active Contours/Snakes Actve Contours/Snakes Erkut Erdem Acknowledgement: The sldes are adapted from the sldes prepared by K. Grauman of Unversty of Texas at Austn Fttng: Edges vs. boundares Edges useful sgnal to ndcate occludng

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

Overview. Basic Setup [9] Motivation and Tasks. Modularization 2008/2/20 IMPROVED COVERAGE CONTROL USING ONLY LOCAL INFORMATION

Overview. Basic Setup [9] Motivation and Tasks. Modularization 2008/2/20 IMPROVED COVERAGE CONTROL USING ONLY LOCAL INFORMATION Overvew 2 IMPROVED COVERAGE CONTROL USING ONLY LOCAL INFORMATION Introducton Mult- Smulator MASIM Theoretcal Work and Smulaton Results Concluson Jay Wagenpfel, Adran Trachte Motvaton and Tasks Basc Setup

More information

A New Image Binarization Method Using Histogram and Spectral Clustering

A New Image Binarization Method Using Histogram and Spectral Clustering A Ne Image Bnarzaton Method Usng Hstogram and Spectral Clusterng Ru Wu 1 Fang Yn Janhua Huang 1 Xanglong Tang 1 1 School of Computer Scence and Technology Harbn Insttute of Technology Harbn Chna School

More information

CS434a/541a: Pattern Recognition Prof. Olga Veksler. Lecture 15

CS434a/541a: Pattern Recognition Prof. Olga Veksler. Lecture 15 CS434a/541a: Pattern Recognton Prof. Olga Veksler Lecture 15 Today New Topc: Unsupervsed Learnng Supervsed vs. unsupervsed learnng Unsupervsed learnng Net Tme: parametrc unsupervsed learnng Today: nonparametrc

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

Learning a Class-Specific Dictionary for Facial Expression Recognition

Learning a Class-Specific Dictionary for Facial Expression Recognition BULGARIAN ACADEMY OF SCIENCES CYBERNETICS AND INFORMATION TECHNOLOGIES Volume 16, No 4 Sofa 016 Prnt ISSN: 1311-970; Onlne ISSN: 1314-4081 DOI: 10.1515/cat-016-0067 Learnng a Class-Specfc Dctonary for

More information

CLUSTERING that discovers the relationship among data

CLUSTERING that discovers the relationship among data Ramp-based Twn Support Vector Clusterng Zhen Wang, Xu Chen, Chun-Na L, and Yuan-Ha Shao arxv:82.0370v [cs.lg] 0 Dec 208 Abstract Tradtonal plane-based clusterng methods measure the cost of wthn-cluster

More information

Optimal Design of Nonlinear Fuzzy Model by Means of Independent Fuzzy Scatter Partition

Optimal Design of Nonlinear Fuzzy Model by Means of Independent Fuzzy Scatter Partition Optmal Desgn of onlnear Fuzzy Model by Means of Independent Fuzzy Scatter Partton Keon-Jun Park, Hyung-Kl Kang and Yong-Kab Km *, Department of Informaton and Communcaton Engneerng, Wonkwang Unversty,

More information

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY 1. SSDH: Semi-supervised Deep Hashing for Large Scale Image Retrieval

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY 1. SSDH: Semi-supervised Deep Hashing for Large Scale Image Retrieval IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY SSDH: Sem-supervsed Deep Hashng for Large Scale Image Retreval Jan Zhang, and Yuxn Peng arxv:607.08477v2 [cs.cv] 8 Jun 207 Abstract Hashng

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

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 Robust Method for Estimating the Fundamental Matrix

A Robust Method for Estimating the Fundamental Matrix Proc. VIIth Dgtal Image Computng: Technques and Applcatons, Sun C., Talbot H., Ourseln S. and Adraansen T. (Eds.), 0- Dec. 003, Sydney A Robust Method for Estmatng the Fundamental Matrx C.L. Feng and Y.S.

More information

Out-of-Sample Extensions for LLE, Isomap, MDS, Eigenmaps, and Spectral Clustering

Out-of-Sample Extensions for LLE, Isomap, MDS, Eigenmaps, and Spectral Clustering Out-of-Sample Extensons for LLE, Isomap, MDS, Egenmaps, and Spectral Clusterng Yoshua Bengo, Jean-Franços Paement, Pascal Vncent Olver Delalleau, Ncolas Le Roux and Mare Oumet Département d Informatque

More information

Concurrent Apriori Data Mining Algorithms

Concurrent Apriori Data Mining Algorithms Concurrent Apror Data Mnng Algorthms Vassl Halatchev Department of Electrcal Engneerng and Computer Scence York Unversty, Toronto October 8, 2015 Outlne Why t s mportant Introducton to Assocaton Rule Mnng

More information

The Discriminate Analysis and Dimension Reduction Methods of High Dimension

The Discriminate Analysis and Dimension Reduction Methods of High Dimension Open Journal of Socal Scences, 015, 3, 7-13 Publshed Onlne March 015 n ScRes. http://www.scrp.org/journal/jss http://dx.do.org/10.436/jss.015.3300 The Dscrmnate Analyss and Dmenson Reducton Methods of

More information

Inverse-Polar Ray Projection for Recovering Projective Transformations

Inverse-Polar Ray Projection for Recovering Projective Transformations nverse-polar Ray Projecton for Recoverng Projectve Transformatons Yun Zhang The Center for Advanced Computer Studes Unversty of Lousana at Lafayette yxz646@lousana.edu Henry Chu The Center for Advanced

More information

Related-Mode Attacks on CTR Encryption Mode

Related-Mode Attacks on CTR Encryption Mode Internatonal Journal of Network Securty, Vol.4, No.3, PP.282 287, May 2007 282 Related-Mode Attacks on CTR Encrypton Mode Dayn Wang, Dongda Ln, and Wenlng Wu (Correspondng author: Dayn Wang) Key Laboratory

More information

Multi-stable Perception. Necker Cube

Multi-stable Perception. Necker Cube Mult-stable Percepton Necker Cube Spnnng dancer lluson, Nobuuk Kaahara Fttng and Algnment Computer Vson Szelsk 6.1 James Has Acknowledgment: Man sldes from Derek Hoem, Lana Lazebnk, and Grauman&Lebe 2008

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

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

Learning-Based Top-N Selection Query Evaluation over Relational Databases

Learning-Based Top-N Selection Query Evaluation over Relational Databases Learnng-Based Top-N Selecton Query Evaluaton over Relatonal Databases Lang Zhu *, Wey Meng ** * School of Mathematcs and Computer Scence, Hebe Unversty, Baodng, Hebe 071002, Chna, zhu@mal.hbu.edu.cn **

More information

Clustering Algorithm Combining CPSO with K-Means Chunqin Gu 1, a, Qian Tao 2, b

Clustering Algorithm Combining CPSO with K-Means Chunqin Gu 1, a, Qian Tao 2, b Internatonal Conference on Advances n Mechancal Engneerng and Industral Informatcs (AMEII 05) Clusterng Algorthm Combnng CPSO wth K-Means Chunqn Gu, a, Qan Tao, b Department of Informaton Scence, Zhongka

More information

Machine Learning 9. week

Machine Learning 9. week Machne Learnng 9. week Mappng Concept Radal Bass Functons (RBF) RBF Networks 1 Mappng It s probably the best scenaro for the classfcaton of two dataset s to separate them lnearly. As you see n the below

More information

Recommended Items Rating Prediction based on RBF Neural Network Optimized by PSO Algorithm

Recommended Items Rating Prediction based on RBF Neural Network Optimized by PSO Algorithm Recommended Items Ratng Predcton based on RBF Neural Network Optmzed by PSO Algorthm Chengfang Tan, Cayn Wang, Yuln L and Xx Q Abstract In order to mtgate the data sparsty and cold-start problems of recommendaton

More information

Maintaining temporal validity of real-time data on non-continuously executing resources

Maintaining temporal validity of real-time data on non-continuously executing resources Mantanng temporal valdty of real-tme data on non-contnuously executng resources Tan Ba, Hong Lu and Juan Yang Hunan Insttute of Scence and Technology, College of Computer Scence, 44, Yueyang, Chna Wuhan

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

Image Alignment CSC 767

Image Alignment CSC 767 Image Algnment CSC 767 Image algnment Image from http://graphcs.cs.cmu.edu/courses/15-463/2010_fall/ Image algnment: Applcatons Panorama sttchng Image algnment: Applcatons Recognton of object nstances

More information

Flatten a Curved Space by Kernel: From Einstein to Euclid

Flatten a Curved Space by Kernel: From Einstein to Euclid Flatten a Curved Space by Kernel: From Ensten to Eucld Quyuan Huang, Dapeng Olver Wu Ensten s general theory of relatvty fundamentally changed our vew about the physcal world. Dfferent from Newton s theory,

More information

Range images. Range image registration. Examples of sampling patterns. Range images and range surfaces

Range images. Range image registration. Examples of sampling patterns. Range images and range surfaces Range mages For many structured lght scanners, the range data forms a hghly regular pattern known as a range mage. he samplng pattern s determned by the specfc scanner. Range mage regstraton 1 Examples

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

Design of Structure Optimization with APDL

Design of Structure Optimization with APDL Desgn of Structure Optmzaton wth APDL Yanyun School of Cvl Engneerng and Archtecture, East Chna Jaotong Unversty Nanchang 330013 Chna Abstract In ths paper, the desgn process of structure optmzaton wth

More information

Efficient Text Classification by Weighted Proximal SVM *

Efficient Text Classification by Weighted Proximal SVM * Effcent ext Classfcaton by Weghted Proxmal SVM * Dong Zhuang 1, Benyu Zhang, Qang Yang 3, Jun Yan 4, Zheng Chen, Yng Chen 1 1 Computer Scence and Engneerng, Bejng Insttute of echnology, Bejng 100081, Chna

More information