A Fast Kernel-based Multilevel Algorithm for Graph Clustering
|
|
- Lucy Lee
- 5 years ago
- Views:
Transcription
1 A Fast Kernel-based Multilevel Algorithm for Graph Clustering Inderjit Dhillon Dept. of Computer Sienes University of Texas at Austin Austin, TX Yuqiang Guan Dept. of Computer Sienes University of Texas at Austin Austin, TX Brian Kulis Dept. of Computer Sienes University of Texas at Austin Austin, TX ABSTRACT Graph lustering (also alled graph partitioning) lustering the nodes of a graph is an important problem in diverse data mining appliations. Traditional approahes involve optimization of graph lustering objetives suh as normalized ut or ratio assoiation; spetral methods are widely used for these objetives, but they require eigenvetor omputation whih an be slow. Reently, graph lustering with a general ut objetive has been shown to be mathematially equivalent to an appropriate weighted kernel k-means objetive funtion. In this paper, we exploit this equivalene to develop a very fast multilevel algorithm for graph lustering. Multilevel approahes involve oarsening, initial partitioning and refinement phases, all of whih may be speialized to different graph lustering objetives. Unlike existing multilevel lustering approahes, suh as METIS, our algorithm does not onstrain the luster sizes to be nearly equal. Our approah gives a theoretial guarantee that the refinement step dereases the graph ut objetive under onsideration. Experiments show that we ahieve better final objetive funtion values as ompared to a state-of-the-art spetral lustering algorithm: on a series of benhmark test graphs with up to thirty thousand nodes and one million edges, our algorithm ahieves lower normalized ut values in 67% of our experiments and higher ratio assoiation values in 100% of our experiments. Furthermore, on large graphs, our algorithm is signifiantly faster than spetral methods. Finally, our algorithm requires far less memory than spetral methods; we luster a 1.2 million node movie network into 5000 lusters, whih due to memory requirements annot be done diretly with spetral methods. Categories and Subjet Desriptors G [Numerial Analysis]: Spetral Methods; H.3.3.a [Information Searh and Retrieval]: Clustering; I.5.3.a [Pattern Reognition]: Clustering Algorithms Permission to make digital or hard opies of all or part of this work for personal or lassroom use is granted without fee provided that opies are not made or distributed for profit or ommerial advantage and that opies bear this notie and the full itation on the first page. To opy otherwise, to republish, to post on servers or to redistribute to lists, requires prior speifi permission and/or a fee. KDD 05, August 21 24, 2005, Chiago, Illinois, USA. Copyright 2005 ACM X/05/ $5.00. General Terms Algorithms, Experimentation Keywords Graph Clustering, Kernel Methods, Multilevel Methods, Spetral Clustering 1. INTRODUCTION Graph lustering (also alled graph partitioning) is an important problem in many domains. Ciruit partitioning, VLSI design, task sheduling, bioinformatis, soial network analysis and a host of other problems all rely on effiient and effetive graph lustering algorithms. In many data mining appliations, pairwise similarities between data objets an be modeled as a graph, and the subsequent problem of data lustering an be viewed as a graph lustering problem. The problem of graph lustering has been studied for deades, and a number of different approahes have been proposed. Spetral methods have been widely used for graph lustering [2, 6, 9]. These algorithms use the eigenvetors of a graph affinity matrix, or a matrix derived from the affinity matrix, to partition the graph. Various spetral algorithms have been developed for a number of different objetive funtions, suh as normalized ut [9] and ratio ut [2]. Furthermore, extensive researh has been done on spetral postproessing: going from the eigenvetor matrix to a disrete lustering [4, 10]. It was reently shown that a wide lass of graph lustering objetives, inluding ratio ut, ratio assoiation, the Kernighan-Lin objetive and normalized ut, an all be viewed as speial ases of the weighted kernel k-means objetive funtion [3, 4]. In addition to unifying several different graph lustering objetives, inluding a number of spetral lustering objetives, this result implies that a simple weighted kernel k-means algorithm an be used to optimize graph lustering objetives. The basi kernel k-means algorithm, however, relies heavily on effetive luster initialization to ahieve good results, and an often yield poor results. In this paper, we develop a multilevel algorithm for graph lustering that uses weighted kernel k-means as the refinement algorithm. Multilevel methods have been used extensively for graph lustering with the Kernighan-Lin objetive, whih attempts to minimize the ut in the graph while maintaining equal-sized lusters [1, 7, 8]. In multilevel algorithms, the graph is repeatedly oarsened level by level until only a small number of nodes are left. Then, an ini-
2 Name Ratio Assoiation Ratio Cut Normalized Cut Objetive Funtion maximize P k links(v,v ) V 1,...,V k =1 V minimize P k links(v,v\v ) V 1,...,V k =1 V minimize P k links(v,v\v ) V 1,...,V k =1 degree(v ) Input Graph Final Clustering Table 1: Examples of graph lustering objetives. Note that Normalized Assoiation = k Normalized Cut. Coarsening Refining tial lustering on this small graph is performed. Finally, the graph is unoarsened level by level, and at eah level, the lustering from the previous level is refined using the refinement algorithm. This multilevel strategy results in a very fast and effetive graph lustering algorithm. METIS [8] is perhaps the most popular multilevel algorithm, and is signifiantly faster than spetral methods on very large graphs when equally sized lusters are desired. However, all previous multilevel algorithms, suh as METIS, suffer from the serious drawbak of restriting lusters to be of equal size. The algorithm presented in this paper removes this restrition: instead of a Kernighan-Lin algorithm for the refinement phase, our algorithm employs weighted kernel k- means. We present experimental results on a number of test graphs to illustrate the advantage of our approah. We show that our algorithm is signifiantly faster than eigenvetor omputation, a neessary step for spetral lustering methods, on large graphs. We also demonstrate that our algorithm outperforms a state-of-the-art spetral lustering algorithm in terms of quality, as measured by the graph objetive funtion values. On our benhmark test graphs of varying sizes, the final objetive funtion values of the multilevel algorithm are better than the spetral algorithm in 100% of runs for ratio assoiation, and 67% of runs for normalized ut. We also present results on data arising from real-life data mining appliations. We luster the 1.2 million node IMDB movie network into 5000 lusters. Running a spetral algorithm diretly on this data set would require storage of 5000 eigenvetors, whih is prohibitive. 2. GRAPH CLUSTERING We first introdue a number different graph lustering objetives. As we will see later, eah of the following objetives may be expressed as speial ases of the weighted kernel k- means objetive funtion, whih will motivate our use of weighted kernel k-means as the algorithm used in the refinement phase. For data given as a graph, a number of different objetives have been proposed for the graph lustering problem. We are given a graph G = (V, E, A), onsisting of a set of verties V and a set of edges E suh that an edge between two verties represents their similarity. The affinity matrix A is V V whose entries represent the weights of the edges (an entry of A is 0 if there is no edge between the orresponding verties). We denote links(a, B) = P i A,j B Aij; that is, the sum of the weights of edges from one luster to the other. Let degree(a) = links(a, V), the links to all verties from nodes in A. A list of some of the most ommon graph lustering obje- Initial Clustering Figure 1: Overview of the multilevel algorithm (for k = 2). tives is presented in Table 1. The Kernighan-Lin objetive, another ommon objetive funtion, is the same as ratio ut exept that lusters are additionally onstrained to be of equal size. For ratio assoiation, ratio ut and normalized ut, the most ommon approah for optimization has been to use a spetral algorithm. Suh an algorithm is derived by relaxing the above objetives so that a global solution to the relaxed objetive may be found by omputing eigenvetors of an appropriate matrix. One these eigenvetors are omputed, a postproessing step derives a disrete lustering from the eigenvetors. See the referenes [2, 9, 10] for further details on the spetral algorithms assoiated with eah of these objetives. For the Kernighan-Lin objetive, the most suessful approah for optimizing the objetive has been to use a multilevel algorithm. 3. THE MULTILEVEL APPROACH In this setion, we develop a new, general algorithm for graph lustering based on multilevel methods. The multilevel approah has been made popular by METIS [8], though a number of multilevel lustering algorithms have been studied, dating bak to [1, 7]. Our algorithm will differ from previous approahes in that it works for a wide lass of graph lustering objetives, and that all three phases of the algorithm an be speialized for eah graph lustering objetive. Below, we desribe eah phase of the algorithm. We assume that we are given an input graph, denoted by G 0 = (V 0, E 0, A 0), as well as the number of partitions requested, k. See Figure 1 for a graphial overview of the multilevel approah. 3.1 Coarsening Phase Given the initial graph G 0, the graph is repeatedly transformed into smaller and smaller graphs G 1, G 2,..., G m suh that V 0 > V 1 >... > V m. To oarsen a graph from G i to G i+1, a number of different tehniques may be used. In general, sets of nodes in G i are ombined to form supernodes in G i+1. When ombining a set of nodes into supernodes, the edge weights between two supernodes in G i+1 are taken to be the sum of the edge weights between the nodes in G i omprising the supernodes.
3 Our oarsening approah works as follows: given a graph, start with all nodes unmarked. Visit eah vertex in a random order. For eah unmarked vertex x, if all neighbors of x have been marked, mark x and proeed to the next unmarked vertex. On the other hand, if x has an unmarked neighbor, then merge x with the unmarked vertex y that maximizes e(x, y) w(x) + e(x, y) w(y), (1) where e(x, y) orresponds to the edge weight between verties x and y and w(x) is the weight of vertex x. Then mark both x and y. One all verties are marked, the oarsening for this level is ompleted. As we will see when disussing the onnetion between graph lustering and weighted kernel k-means, different graph lustering objetives indue different vertex weights. For example, in normalized ut, the weight of a vertex is its degree, and (1) redues to the normalized ut between x and y. For ratio assoiation, (1) simplifies to the heaviest edge riterion of METIS, sine all verties have unit weight. 3.2 Initial Clustering Phase Eventually, the graph is oarsened to the point where very few nodes remain in the graph. We speify a parameter indiating how small we want the oarsest graph to be; in our experiments, we stop oarsening when the graph has less than 20k nodes, where k is the number of desired lusters. At this point, we perform an initial partitioning by lustering the oarsest graph. One way to obtain an initial partitioning is to use the region growing algorithm of METIS [8]. However, this approah is inadequate for our purposes sine it generates lusters of equal size. We have found the best method for initialization to be a spetral algorithm. We use the spetral algorithm of Yu and Shi [10], whih we generalize to work with arbitrary weights [4]. Thus our initial lustering is ustomized for different graph lustering objetives. Sine the oarsest graph is signifiantly smaller than the input graph, spetral methods are adequate in terms of speed. We find the spetral methods to yield the best quality of results, though they are somewhat less effiient than the region growing algorithm. Hene, when effiieny is a onern, the region growing algorithm may be used instead. 3.3 Refinement Phase The final phase of the algorithm is the refinement phase. Given a graph G i, we form the graph G i 1 (G i 1 is the same graph used in level i 1 in the oarsening phase). We extend the lustering from G i to G i 1 as follows: if a supernode is in a luster, then all nodes in G i formed from that supernode are in luster. This yields an initial lustering for the graph, whih is then improved using a refinement algorithm. Note that we also run the refinement algorithm on the oarsest graph. The algorithm terminates after the refinement algorithm is run on the original graph G 0. Sine we have a good initial lustering at eah level, the refinement algorithm often onverges very quikly. Thus, this approah is extremely effiient. At eah refinement step, our algorithm uses an approah inspired by the reently-shown theoretial onnetion between kernel k-means and spetral lustering [4]. In [4], we proved that eah of the graph lustering objetives given Algorithm 1: Refinement Algorithm. Weighted Kernel kmeans(k, k, w, t max, {π (0) } k =1, {π } k =1) Input: K: kernel matrix, k: number of lusters, w: weights for eah point, t max: optional maximum number of iterations, {π (0) } k =1: optional initial lustering Output: {π } k =1: final lustering of the points 1. If no initial lustering is given, initialize the k lusters π (0) 1,..., π(0) k randomly. Set t = For eah row i of K and every luster, ompute d(i,m ) = K ii 2 P j π (t) w jk ij P + P j,l π (t) ( P j π (t) j π (t) w j w jw l K jl. w j) 2 3. Find (i) = argmin d(i,m ), resolving ties arbitrarily. Compute the updated lusters as π (t+1) = {i : (i) = }. 4. If not onverged or t max > t, set t = t + 1 and go to Step 3; Otherwise, stop and output final lusters {π (t+1) } k =1. Objetive Node Weights Kernel Matrix Ratio Assoiation 1 nodes K = σi + A Ratio Cut 1 nodes K = σi D + A K-L Objetive 1 nodes K = σi D + A Normalized Cut Deg. of node K = σd 1 + D 1 AD 1 Table 2: Popular graph lustering objetives with orresponding weights and kernels given the affinity matrix A earlier may be expressed as speial ases of the weighted kernel k-means objetive with the orret hoie of weights and kernel matrix. Hene, the weighted kernel k-means algorithm an be diretly used to loally optimize these graph lustering objetives. Eah iteration of this algorithm osts O(nz) time, where nz is the number of nonzero entries in the kernel matrix. Algorithm 1 desribes the weighted kernel k-means algorithm. The input is the kernel matrix K, the number of lusters k, weights for the points w, an optional maximum number of iterations, and an optional initial lustering, denoted by {π (0) } k =1. For the Kernighan-Lin (K-L) objetive, whih requires equally sized lusters, an inremental kernel k-means algorithm (in whih points are swapped to improve the objetive funtion value) an be used to loally optimize the objetive. In Table 2, we display eah of the earlier graph lustering objetives, along with the hoie of weights and kernel required for the weighted kernel k-means algorithm to monotonially optimize the given graph lustering objetive. The matrix D is a diagonal matrix whose entries orrespond to the sum of the rows of A, the graph affinity matrix. Finally, σ is a real number hosen to be large enough that K is positive definite. As long as K is positive definite, we
4 Graph name #nodes #edges Desription DATA finite element mesh 3ELT finite element mesh UK finite element mesh ADD bit adder WHITAKER finite element mesh CRACK finite element mesh FE 4ELT finite element mesh MEMPLUS memory iruit BCSSTK stiffness matrix kkm METIS NCV RAV Table 4: Normalized ut values (NCV) and ratio assoiation values (RAV) for obtaining 5000 lusters of the IMDB movie data set using our multilevel algorithm (), kernel k-means (kkm) and METIS Table 3: Test graphs an theoretially guarantee that the algorithm monotonially optimizes the orresponding graph lustering objetive. Initialization an be done randomly, or by using another lustering algorithm, suh as METIS or spetral lustering. In the ase of our multilevel algorithm, initialization is obtained from the lustering from the previous level, or the initial lustering in the ase of the oarsest graph. An extension to the basi bath algorithm presented in [4] uses loal searh to avoid loal optima. To avoid this problem, it has been shown [5] that loal searh an signifiantly improve results by allowing the algorithm to esape suh poor optima. The loal searh algorithm onsiders the effet on the objetive funtion of moving a point from one luster to another. It hooses a hain of moves that auses the greatest improvement in the objetive funtion value. We inorporate this loal searh proedure to improve the quality of our weighted kernel k-means bath refinement sheme. 4. EXPERIMENTS In this setion, we present a number of experiments to show that our multilevel weighted kernel k-means algorithm outperforms spetral methods in terms of quality (normalized ut and ratio assoiation values) as well as omputation time. We also present results of lustering the 1.2 million node IMDB movie data set graph. In our experiments, we use the spetral algorithm from [10], generalized to work for both normalized ut and ratio assoiation. This algorithm is used as the benhmark spetral algorithm in our experiments, as well as the method for lustering the oarsest graph in our multilevel implementation. To speed up omputation, we onsidered only boundary points when running weighted kernel k-means; further disussion is omitted due to lak of spae. All the experiments are performed on a Linux mahine with 2.4 GHz Pentium IV proessor and 1 GB main memory. Additionally, we use the following aronyms: stands for our multilevel kernel k-means algorithm with no loal searh, is our multilevel algorithm with a loal searh hain length of 20, kkm stands for kernel k- means with random initialization, NCV is short for normalized ut value, and RAV is short for ratio assoiation value. 4.1 Clustering of Benhmark Graphs Table 3 lists 9 test graphs 1 from various soures and different appliation domains. Some of them are benhmark 1 Downloaded from partition/ matries used to test METIS. On eah of the benhmark graphs, we ompared our multilevel algorithms, and, with the spetral algorithm in terms of objetive funtion values and omputation time. We report results for both ratio assoiation and normalized ut, and when 64 lusters are omputed. We do not report omparisons between our multilevel algorithm and METIS; in [4], we showed that the spetral algorithm is superior to METIS in terms of better ratio assoiation and normalized ut values. Figure 2 shows the relative performane of the spetral algorithm ompared with our multilevel algorithm in terms of RAV and omputation time. In the left panel, for eah graph we plot the ratio of the RAV of all methods to that of ratios less than 1 indiate that produes higher RAVs (and thus performs better). We an see that 100% of the RAVs produed using are higher than those obtained by the spetral algorithm. Also, we see that performs onsistently better than, whih implies that loal searh improves the RAV in all ases (maximally by 3%). In the right panel of Figure 2, we plot the omputation time for the three methods. It is lear that our multilevel algorithm is muh faster than the spetral algorithm. For example, in the ase of ADD32, the spetral algorithm is 48 times slower than and 55 times slower than. Results for the normalized ut objetive appear in Figure 3. For eah graph, we plot the ratio of NCV of other methods to that of. As opposed to ratio assoiation, values that are larger than 1 in Figure 3 indiate that produes lower NCVs (and thus performs better). We see that 67% of the NCVs produed using are lower than those produed using the spetral algorithm. We also see that, whih utilizes the loal searh tehnique, performs muh better than in many ases. In terms of omputation time, our multilevel methods again outperform the spetral algorithm in all ases. For ADD32, spetral is 58 times slower than and 60 times slower than. 4.2 The IMDB Movie Data Set A Case Study The IMDB data set 2 ontains information about movies, musi bands, ators, movie festivals and events. By onneting ators and movies or events in whih they partiipate, we form a sparse undireted graph with approximately 1.2 million nodes and 7.6 million edges. It is impratial to run a spetral algorithm diretly on this data set to produe 5000 lusters not only beause om- 2 Downloaded from
5 Ratio assoiation value (saled by values of ) spetral 64 lusters Computation time in seonds spetral 64 lusters 0.9 data 3elt uk add32 whitaker3 rak fe_4elt2 memplus bsstk30 0 data 3elt uk add32 whitaker3 rak fe_4elt2 memplus bsstk30 Figure 2: Quality and omputation time of our multilevel methods ompared with the benhmark spetral algorithm. The left panel plots the ratio of the ratio assoiation value of every algorithm to that of for 64 lusters. Note that bars below the baseline orrespond to ases where performs better. The right panel ompares omputation times for generating 64 lusters using different algorithms. As an example, on the ADD32 graph, the spetral algorithm is 48 times slower than and 58 times slower than. Normalized ut value (saled by values of ) spetral 64 lusters Computation time in seonds spetral 64 lusters 0.9 data 3elt uk add32 whitaker3 rak fe_4elt2 memplus bsstk30 0 data 3elt uk add32 whitaker3 rak fe_4elt2 memplus bsstk30 Figure 3: Quality and omputation time of our multilevel methods ompared with the benhmark spetral algorithm. The left panel plots the ratio of the normalized ut value of every algorithm to that of for 64 lusters. Note that bars above the baseline orrespond to the ases where performs better. The right panel ompares omputation times for generating 64 lusters. As an example, on the ADD32 graph, the spetral algorithm is 58 times slower than and 60 times slower than. Movies Harry Potter and the Sorerer s Stone (2001) Harry Potter and the Chamber of Serets (2002) Harry Potter and the Prisoner of Azkaban (2004) Harry Potter and the Goblet of Fire (2005) Harry Potter: Behind the Magi (2001 TV) Harry Potter und die Kammer des Shrekens: Das grobe RTL Speial zum Film (2002 TV) Ators Daniel Radliffe, Rupert Grint, Emma Watson, Tom Felton Peter Best, Sean Biggerstaff, Sott Fern, Alfred Enoh, Joshua Herdman Harry Melling, Matthew Lewis, Devon Murray, Robert Pattinson James Phelps, Oliver Phelps, Edward Randell, Jamie Waylett Shefali Chowdhury, Katie Leung, Bonnie Wright, Stanislav Ianevski Jamie Yeates, Chris Rankin Table 5: A Seletion of Movies and Ators in Cluster 633
6 nz = nz = Figure 4: Graphial Plots of Cluster 633 of IMDB (Harry Potter). The right plot shows a loser view of the luster irled in the left plot. puting 5000 eigenvetors of a graph of this size would be extremely slow but also beause storing 5000 eigenvetors in main memory requires 24 GB, whih is prohibitive. Thus our multilevel algorithm is also onsiderably more effiient in terms of memory usage in addition to omputation time. Sine we annot run spetral methods on this data set, we ompare our multilevel algorithm () to METIS and weighted kernel k-means with random initialization (kkm). It takes our multilevel algorithm approximately twenty-five minutes to luster this graph into 5000 lusters, and the luster sizes range from 13 to 7616 (for the normalized ut objetive). Table 4 shows the normalized ut values and ratio assoiation values for 5000 lusters generated using these three methods. We see that our multilevel algorithm is markedly superior to kkm: its NCV is twie as high as the NCV of and the RAV of kkm is less than one tenth of the RAV of. Comparing METIS and, the latter is superior and ahieves a RAV that is 50% higher and a NCV that is 10% lower. To demonstrate the quality of results, we disuss a sampling of the produed lusters. After lustering, we rearrange the rows and olumns suh that rows (olumns) in the same luster are adjaent to one another. Cluster 633, irled in the right plot of Figure 4, is of size 121 and mainly ontains Harry Potter movies and the ators in these movies. The left olumn of Table 5 lists movies in the luster, where we see 3 Harry Potter movies that were released in the past and 1 to be released in November, There are also 2 other doumentary TV programs. The right olumn of Table 5 lists a seletion of some of the ators in the luster, where we see the major ast members of the Harry Potter movies, suh as Daniel Radliffe, Rupert Grint, Emma Watson, et. Cluster 3537 ontains the short series Festival número (No. 1-12), shot in year 1965; luster 4400 ontains the Korean movie Chunhyang and 19 of its ast members who ated only in this movie in the database. Other small lusters follow this pattern, i.e., most of them are about one movie or movie series and ontain ast members that ated only in this movie or movie series. Popular ators, diretors or well-known movie festivals are assoiated with more people, so generally they belong to muh larger lusters. 5. CONCLUSIONS We have presented a new multilevel algorithm for graph lustering. This algorithm has key benefits over existing graph lustering algorithms. Unlike previous multilevel algorithms suh as METIS, our algorithm is general and aptures a number of graph lustering objetives. This frees us from the restrition of equal-size lusters. It has advantages over spetral methods in that the multilevel approah is onsiderably faster and gives better final objetive funtion values. Using a number of benhmark test graphs, we demonstrated that, in general, our algorithm produes better ratio assoiation values and normalized ut values when using spetral initialization at the oarsest level. As the oarsest graph is signifiantly smaller than the input graph, spetral methods are feasible in this ase, and the running time of the algorithm is still muh faster than spetral lustering on the input graph. Spetral methods have attrated intense study over the last several years as a powerful method for graph lustering; our results indiate that this may not be the most powerful tehnique for partitioning graphs. Furthermore, given an input pairwise similarity matrix between data vetors, our approah an be viewed as a fast multilevel algorithm for optimizing the kernel k-means objetive. Our approah also onsumes far less memory than spetral methods. Spetral lustering of a 1.2 million node movie network into 5000 lusters would require several gigabytes of storage. On the other hand, we were able to luster this network in approximately twenty-five minutes without any signifiant memory overhead. Aknowledgments. This researh was supported by NSF grant CCF , NSF Career Award ACI , and NSF-ITR award IIS REFERENCES [1] T. Bui and C. Jones. A heuristi for reduing fill-in in sparse matrix fatorization. In 6th SIAM Conferene on Parallel Proessing for Sientifi Computing, pages , [2] P. Chan, M. Shlag, and J. Zien. Spetral k-way ratio ut partitioning. IEEE Trans. CAD-Integrated Ciruits and Systems, 13: , [3] I. Dhillon, Y. Guan, and B. Kulis. Kernel k-means, spetral lustering and normalized uts. In Pro. 10th ACM KDD Conferene, [4] I. Dhillon, Y. Guan, and B. Kulis. A unified view of kernel k-means, spetral lustering and graph uts. Tehnial Report TR-04-25, University of Texas at Austin, [5] I. S. Dhillon, Y. Guan, and J. Kogan. Iterative lustering of high dimensional text data augmented by loal searh. In Proeedings of The 2002 IEEE International Conferene on Data Mining, pages , [6] W. E. Donath and A. J. Hoffman. Lower bounds for the partitioning of graphs. IBM J. Res. Development, 17: , [7] B. Hendrikson and R. Leland. A multilevel algorithm for partitioning graphs. Tehnial Report SAND , Sandia National Laboratories, [8] G. Karypis and V. Kumar. A fast and high quality multilevel sheme for partitioning irregular graphs. SIAM J. Si. Comput., 20(1): , [9] J. Shi and J. Malik. Normalized uts and image segmentation. IEEE Trans. Pattern Analysis and Mahine Intelligene, 22(8): , August [10] S. X. Yu and J. Shi. Multilass spetral lustering. In International Conferene on Computer Vision, 2003.
Normalized Cuts Without Eigenvectors: A Multilevel Approach
Normalized Cuts Without Eigenvectors: A Multilevel Approach Inderjit S. Dhillon The University of Texas at Austin SIAM Conference on Parallel Processing February 24, 2006 Joint work with Yuqiang Guan and
More informationGraph Clustering and Co-clustering
Inderjit S. Dhillon University of Texas at Austin IPAM Nov 5, 2007 Joint work with A. Banerjee, J. Ghosh, Y. Guan, B. Kulis, S. Merugu & D. Modha Outline Clustering Graphs: Spectral Clustering & A Surprising
More informationThe Minimum Redundancy Maximum Relevance Approach to Building Sparse Support Vector Machines
The Minimum Redundany Maximum Relevane Approah to Building Sparse Support Vetor Mahines Xiaoxing Yang, Ke Tang, and Xin Yao, Nature Inspired Computation and Appliations Laboratory (NICAL), Shool of Computer
More informationNONLINEAR BACK PROJECTION FOR TOMOGRAPHIC IMAGE RECONSTRUCTION. Ken Sauer and Charles A. Bouman
NONLINEAR BACK PROJECTION FOR TOMOGRAPHIC IMAGE RECONSTRUCTION Ken Sauer and Charles A. Bouman Department of Eletrial Engineering, University of Notre Dame Notre Dame, IN 46556, (219) 631-6999 Shool of
More informationthe data. Structured Principal Component Analysis (SPCA)
Strutured Prinipal Component Analysis Kristin M. Branson and Sameer Agarwal Department of Computer Siene and Engineering University of California, San Diego La Jolla, CA 9193-114 Abstrat Many tasks involving
More informationA Novel Validity Index for Determination of the Optimal Number of Clusters
IEICE TRANS. INF. & SYST., VOL.E84 D, NO.2 FEBRUARY 2001 281 LETTER A Novel Validity Index for Determination of the Optimal Number of Clusters Do-Jong KIM, Yong-Woon PARK, and Dong-Jo PARK, Nonmembers
More informationAutomatic Physical Design Tuning: Workload as a Sequence Sanjay Agrawal Microsoft Research One Microsoft Way Redmond, WA, USA +1-(425)
Automati Physial Design Tuning: Workload as a Sequene Sanjay Agrawal Mirosoft Researh One Mirosoft Way Redmond, WA, USA +1-(425) 75-357 sagrawal@mirosoft.om Eri Chu * Computer Sienes Department University
More informationPipelined Multipliers for Reconfigurable Hardware
Pipelined Multipliers for Reonfigurable Hardware Mithell J. Myjak and José G. Delgado-Frias Shool of Eletrial Engineering and Computer Siene, Washington State University Pullman, WA 99164-2752 USA {mmyjak,
More informationExtracting Partition Statistics from Semistructured Data
Extrating Partition Statistis from Semistrutured Data John N. Wilson Rihard Gourlay Robert Japp Mathias Neumüller Department of Computer and Information Sienes University of Strathlyde, Glasgow, UK {jnw,rsg,rpj,mathias}@is.strath.a.uk
More informationA Unified View of Kernel k-means, Spectral Clustering and Graph Cuts
A Unified View of Kernel k-means, Spectral Clustering and Graph Cuts Inderjit Dhillon, Yuqiang Guan and Brian Kulis University of Texas at Austin Department of Computer Sciences Austin, TX 78712 UTCS Technical
More informationA {k, n}-secret Sharing Scheme for Color Images
A {k, n}-seret Sharing Sheme for Color Images Rastislav Luka, Konstantinos N. Plataniotis, and Anastasios N. Venetsanopoulos The Edward S. Rogers Sr. Dept. of Eletrial and Computer Engineering, University
More informationCleanUp: Improving Quadrilateral Finite Element Meshes
CleanUp: Improving Quadrilateral Finite Element Meshes Paul Kinney MD-10 ECC P.O. Box 203 Ford Motor Company Dearborn, MI. 8121 (313) 28-1228 pkinney@ford.om Abstrat: Unless an all quadrilateral (quad)
More informationINTERPOLATED AND WARPED 2-D DIGITAL WAVEGUIDE MESH ALGORITHMS
Proeedings of the COST G-6 Conferene on Digital Audio Effets (DAFX-), Verona, Italy, Deember 7-9, INTERPOLATED AND WARPED -D DIGITAL WAVEGUIDE MESH ALGORITHMS Vesa Välimäki Lab. of Aoustis and Audio Signal
More informationAn Optimized Approach on Applying Genetic Algorithm to Adaptive Cluster Validity Index
IJCSES International Journal of Computer Sienes and Engineering Systems, ol., No.4, Otober 2007 CSES International 2007 ISSN 0973-4406 253 An Optimized Approah on Applying Geneti Algorithm to Adaptive
More informationAnalysis of input and output configurations for use in four-valued CCD programmable logic arrays
nalysis of input and output onfigurations for use in four-valued D programmable logi arrays J.T. utler H.G. Kerkhoff ndexing terms: Logi, iruit theory and design, harge-oupled devies bstrat: s in binary,
More informationHEXA: Compact Data Structures for Faster Packet Processing
Washington University in St. Louis Washington University Open Sholarship All Computer Siene and Engineering Researh Computer Siene and Engineering Report Number: 27-26 27 HEXA: Compat Data Strutures for
More informationarxiv: v1 [cs.db] 13 Sep 2017
An effiient lustering algorithm from the measure of loal Gaussian distribution Yuan-Yen Tai (Dated: May 27, 2018) In this paper, I will introdue a fast and novel lustering algorithm based on Gaussian distribution
More informationExploiting Enriched Contextual Information for Mobile App Classification
Exploiting Enrihed Contextual Information for Mobile App Classifiation Hengshu Zhu 1 Huanhuan Cao 2 Enhong Chen 1 Hui Xiong 3 Jilei Tian 2 1 University of Siene and Tehnology of China 2 Nokia Researh Center
More informationAdapting K-Medians to Generate Normalized Cluster Centers
Adapting -Medians to Generate Normalized Cluster Centers Benamin J. Anderson, Deborah S. Gross, David R. Musiant Anna M. Ritz, Thomas G. Smith, Leah E. Steinberg Carleton College andersbe@gmail.om, {dgross,
More informationApproximate logic synthesis for error tolerant applications
Approximate logi synthesis for error tolerant appliations Doohul Shin and Sandeep K. Gupta Eletrial Engineering Department, University of Southern California, Los Angeles, CA 989 {doohuls, sandeep}@us.edu
More informationLearning Convention Propagation in BeerAdvocate Reviews from a etwork Perspective. Abstract
CS 9 Projet Final Report: Learning Convention Propagation in BeerAdvoate Reviews from a etwork Perspetive Abstrat We look at the way onventions propagate between reviews on the BeerAdvoate dataset, and
More informationAbstract. Key Words: Image Filters, Fuzzy Filters, Order Statistics Filters, Rank Ordered Mean Filters, Channel Noise. 1.
Fuzzy Weighted Rank Ordered Mean (FWROM) Filters for Mixed Noise Suppression from Images S. Meher, G. Panda, B. Majhi 3, M.R. Meher 4,,4 Department of Eletronis and I.E., National Institute of Tehnology,
More informationCluster-Based Cumulative Ensembles
Cluster-Based Cumulative Ensembles Hanan G. Ayad and Mohamed S. Kamel Pattern Analysis and Mahine Intelligene Lab, Eletrial and Computer Engineering, University of Waterloo, Waterloo, Ontario N2L 3G1,
More informationMulti-Piece Mold Design Based on Linear Mixed-Integer Program Toward Guaranteed Optimality
INTERNATIONAL CONFERENCE ON MANUFACTURING AUTOMATION (ICMA200) Multi-Piee Mold Design Based on Linear Mixed-Integer Program Toward Guaranteed Optimality Stephen Stoyan, Yong Chen* Epstein Department of
More informationOn - Line Path Delay Fault Testing of Omega MINs M. Bellos 1, E. Kalligeros 1, D. Nikolos 1,2 & H. T. Vergos 1,2
On - Line Path Delay Fault Testing of Omega MINs M. Bellos, E. Kalligeros, D. Nikolos,2 & H. T. Vergos,2 Dept. of Computer Engineering and Informatis 2 Computer Tehnology Institute University of Patras,
More informationBoosted Random Forest
Boosted Random Forest Yohei Mishina, Masamitsu suhiya and Hironobu Fujiyoshi Department of Computer Siene, Chubu University, 1200 Matsumoto-ho, Kasugai, Aihi, Japan {mishi, mtdoll}@vision.s.hubu.a.jp,
More informationImproved Circuit-to-CNF Transformation for SAT-based ATPG
Improved Ciruit-to-CNF Transformation for SAT-based ATPG Daniel Tille 1 René Krenz-Bååth 2 Juergen Shloeffel 2 Rolf Drehsler 1 1 Institute of Computer Siene, University of Bremen, 28359 Bremen, Germany
More informationA DYNAMIC ACCESS CONTROL WITH BINARY KEY-PAIR
Malaysian Journal of Computer Siene, Vol 10 No 1, June 1997, pp 36-41 A DYNAMIC ACCESS CONTROL WITH BINARY KEY-PAIR Md Rafiqul Islam, Harihodin Selamat and Mohd Noor Md Sap Faulty of Computer Siene and
More informationKERNEL SPARSE REPRESENTATION WITH LOCAL PATTERNS FOR FACE RECOGNITION
KERNEL SPARSE REPRESENTATION WITH LOCAL PATTERNS FOR FACE RECOGNITION Cuiui Kang 1, Shengai Liao, Shiming Xiang 1, Chunhong Pan 1 1 National Laboratory of Pattern Reognition, Institute of Automation, Chinese
More informationA Novel Bit Level Time Series Representation with Implication of Similarity Search and Clustering
A Novel Bit Level Time Series Representation with Impliation of Similarity Searh and lustering hotirat Ratanamahatana, Eamonn Keogh, Anthony J. Bagnall 2, and Stefano Lonardi Dept. of omputer Siene & Engineering,
More informationAnd, the (low-pass) Butterworth filter of order m is given in the frequency domain by
Problem Set no.3.a) The ideal low-pass filter is given in the frequeny domain by B ideal ( f ), f f; =, f > f. () And, the (low-pass) Butterworth filter of order m is given in the frequeny domain by B
More informationUsing Augmented Measurements to Improve the Convergence of ICP
Using Augmented Measurements to Improve the onvergene of IP Jaopo Serafin, Giorgio Grisetti Dept. of omputer, ontrol and Management Engineering, Sapienza University of Rome, Via Ariosto 25, I-0085, Rome,
More informationExploring the Commonality in Feature Modeling Notations
Exploring the Commonality in Feature Modeling Notations Miloslav ŠÍPKA Slovak University of Tehnology Faulty of Informatis and Information Tehnologies Ilkovičova 3, 842 16 Bratislava, Slovakia miloslav.sipka@gmail.om
More informationUnsupervised Stereoscopic Video Object Segmentation Based on Active Contours and Retrainable Neural Networks
Unsupervised Stereosopi Video Objet Segmentation Based on Ative Contours and Retrainable Neural Networks KLIMIS NTALIANIS, ANASTASIOS DOULAMIS, and NIKOLAOS DOULAMIS National Tehnial University of Athens
More informationAlgorithms for External Memory Lecture 6 Graph Algorithms - Weighted List Ranking
Algorithms for External Memory Leture 6 Graph Algorithms - Weighted List Ranking Leturer: Nodari Sithinava Sribe: Andi Hellmund, Simon Ohsenreither 1 Introdution & Motivation After talking about I/O-effiient
More informationA Partial Sorting Algorithm in Multi-Hop Wireless Sensor Networks
A Partial Sorting Algorithm in Multi-Hop Wireless Sensor Networks Abouberine Ould Cheikhna Department of Computer Siene University of Piardie Jules Verne 80039 Amiens Frane Ould.heikhna.abouberine @u-piardie.fr
More informationSparse Certificates for 2-Connectivity in Directed Graphs
Sparse Certifiates for 2-Connetivity in Direted Graphs Loukas Georgiadis Giuseppe F. Italiano Aikaterini Karanasiou Charis Papadopoulos Nikos Parotsidis Abstrat Motivated by the emergene of large-sale
More informationCalculation of typical running time of a branch-and-bound algorithm for the vertex-cover problem
Calulation of typial running time of a branh-and-bound algorithm for the vertex-over problem Joni Pajarinen, Joni.Pajarinen@iki.fi Otober 21, 2007 1 Introdution The vertex-over problem is one of a olletion
More informationGray Codes for Reflectable Languages
Gray Codes for Refletable Languages Yue Li Joe Sawada Marh 8, 2008 Abstrat We lassify a type of language alled a refletable language. We then develop a generi algorithm that an be used to list all strings
More informationAlleviating DFT cost using testability driven HLS
Alleviating DFT ost using testability driven HLS M.L.Flottes, R.Pires, B.Rouzeyre Laboratoire d Informatique, de Robotique et de Miroéletronique de Montpellier, U.M. CNRS 5506 6 rue Ada, 34392 Montpellier
More information13.1 Numerical Evaluation of Integrals Over One Dimension
13.1 Numerial Evaluation of Integrals Over One Dimension A. Purpose This olletion of subprograms estimates the value of the integral b a f(x) dx where the integrand f(x) and the limits a and b are supplied
More informationThe AMDREL Project in Retrospective
The AMDREL Projet in Retrospetive K. Siozios 1, G. Koutroumpezis 1, K. Tatas 1, N. Vassiliadis 2, V. Kalenteridis 2, H. Pournara 2, I. Pappas 2, D. Soudris 1, S. Nikolaidis 2, S. Siskos 2, and A. Thanailakis
More informationProblem Definition. Clustering nonlinearly separable data:
Outlines Weighted Graph Cuts without Eigenvectors: A Multilevel Approach (PAMI 2007) User-Guided Large Attributed Graph Clustering with Multiple Sparse Annotations (PAKDD 2016) Problem Definition Clustering
More informationGradient based progressive probabilistic Hough transform
Gradient based progressive probabilisti Hough transform C.Galambos, J.Kittler and J.Matas Abstrat: The authors look at the benefits of exploiting gradient information to enhane the progressive probabilisti
More informationReduced-Complexity Column-Layered Decoding and. Implementation for LDPC Codes
Redued-Complexity Column-Layered Deoding and Implementation for LDPC Codes Zhiqiang Cui 1, Zhongfeng Wang 2, Senior Member, IEEE, and Xinmiao Zhang 3 1 Qualomm In., San Diego, CA 92121, USA 2 Broadom Corp.,
More informationSystem-Level Parallelism and Throughput Optimization in Designing Reconfigurable Computing Applications
System-Level Parallelism and hroughput Optimization in Designing Reonfigurable Computing Appliations Esam El-Araby 1, Mohamed aher 1, Kris Gaj 2, arek El-Ghazawi 1, David Caliga 3, and Nikitas Alexandridis
More informationSequential Incremental-Value Auctions
Sequential Inremental-Value Autions Xiaoming Zheng and Sven Koenig Department of Computer Siene University of Southern California Los Angeles, CA 90089-0781 {xiaominz,skoenig}@us.edu Abstrat We study the
More information特集 Road Border Recognition Using FIR Images and LIDAR Signal Processing
デンソーテクニカルレビュー Vol. 15 2010 特集 Road Border Reognition Using FIR Images and LIDAR Signal Proessing 高木聖和 バーゼル ファルディ Kiyokazu TAKAGI Basel Fardi ヘンドリック ヴァイゲル Hendrik Weigel ゲルド ヴァニーリック Gerd Wanielik This paper
More informationFUZZY WATERSHED FOR IMAGE SEGMENTATION
FUZZY WATERSHED FOR IMAGE SEGMENTATION Ramón Moreno, Manuel Graña Computational Intelligene Group, Universidad del País Vaso, Spain http://www.ehu.es/winto; {ramon.moreno,manuel.grana}@ehu.es Abstrat The
More informationOne Against One or One Against All : Which One is Better for Handwriting Recognition with SVMs?
One Against One or One Against All : Whih One is Better for Handwriting Reognition with SVMs? Jonathan Milgram, Mohamed Cheriet, Robert Sabourin To ite this version: Jonathan Milgram, Mohamed Cheriet,
More informationAutomated System for the Study of Environmental Loads Applied to Production Risers Dustin M. Brandt 1, Celso K. Morooka 2, Ivan R.
EngOpt 2008 - International Conferene on Engineering Optimization Rio de Janeiro, Brazil, 01-05 June 2008. Automated System for the Study of Environmental Loads Applied to Prodution Risers Dustin M. Brandt
More informationSelf-Adaptive Parent to Mean-Centric Recombination for Real-Parameter Optimization
Self-Adaptive Parent to Mean-Centri Reombination for Real-Parameter Optimization Kalyanmoy Deb and Himanshu Jain Department of Mehanial Engineering Indian Institute of Tehnology Kanpur Kanpur, PIN 86 {deb,hjain}@iitk.a.in
More informationPerformance of Histogram-Based Skin Colour Segmentation for Arms Detection in Human Motion Analysis Application
World Aademy of Siene, Engineering and Tehnology 8 009 Performane of Histogram-Based Skin Colour Segmentation for Arms Detetion in Human Motion Analysis Appliation Rosalyn R. Porle, Ali Chekima, Farrah
More informationThe Implementation of RRTs for a Remote-Controlled Mobile Robot
ICCAS5 June -5, KINEX, Gyeonggi-Do, Korea he Implementation of RRs for a Remote-Controlled Mobile Robot Chi-Won Roh*, Woo-Sub Lee **, Sung-Chul Kang *** and Kwang-Won Lee **** * Intelligent Robotis Researh
More informationVolume 3, Issue 9, September 2013 International Journal of Advanced Research in Computer Science and Software Engineering
Volume 3, Issue 9, September 2013 ISSN: 2277 128X International Journal of Advaned Researh in Computer Siene and Software Engineering Researh Paper Available online at: www.ijarsse.om A New-Fangled Algorithm
More informationAn Edge-based Clustering Algorithm to Detect Social Circles in Ego Networks
JOURNAL OF COMPUTERS, VOL. 8, NO., OCTOBER 23 2575 An Edge-based Clustering Algorithm to Detet Soial Cirles in Ego Networks Yu Wang Shool of Computer Siene and Tehnology, Xidian University Xi an,77, China
More informationThis fact makes it difficult to evaluate the cost function to be minimized
RSOURC LLOCTION N SSINMNT In the resoure alloation step the amount of resoures required to exeute the different types of proesses is determined. We will refer to the time interval during whih a proess
More informationA Load-Balanced Clustering Protocol for Hierarchical Wireless Sensor Networks
International Journal of Advanes in Computer Networks and Its Seurity IJCNS A Load-Balaned Clustering Protool for Hierarhial Wireless Sensor Networks Mehdi Tarhani, Yousef S. Kavian, Saman Siavoshi, Ali
More informationDirected Rectangle-Visibility Graphs have. Abstract. Visibility representations of graphs map vertices to sets in Euclidean space and
Direted Retangle-Visibility Graphs have Unbounded Dimension Kathleen Romanik DIMACS Center for Disrete Mathematis and Theoretial Computer Siene Rutgers, The State University of New Jersey P.O. Box 1179,
More informationCOMBINATION OF INTERSECTION- AND SWEPT-BASED METHODS FOR SINGLE-MATERIAL REMAP
Combination of intersetion- and swept-based methods for single-material remap 11th World Congress on Computational Mehanis WCCM XI) 5th European Conferene on Computational Mehanis ECCM V) 6th European
More informationSemi-Supervised Affinity Propagation with Instance-Level Constraints
Semi-Supervised Affinity Propagation with Instane-Level Constraints Inmar E. Givoni, Brendan J. Frey Probabilisti and Statistial Inferene Group University of Toronto 10 King s College Road, Toronto, Ontario,
More informationA Dual-Hamiltonian-Path-Based Multicasting Strategy for Wormhole-Routed Star Graph Interconnection Networks
A Dual-Hamiltonian-Path-Based Multiasting Strategy for Wormhole-Routed Star Graph Interonnetion Networks Nen-Chung Wang Department of Information and Communiation Engineering Chaoyang University of Tehnology,
More informationDrawing lines. Naïve line drawing algorithm. drawpixel(x, round(y)); double dy = y1 - y0; double dx = x1 - x0; double m = dy / dx; double y = y0;
Naïve line drawing algorithm // Connet to grid points(x0,y0) and // (x1,y1) by a line. void drawline(int x0, int y0, int x1, int y1) { int x; double dy = y1 - y0; double dx = x1 - x0; double m = dy / dx;
More informationOn the Generation of Multiplexer Circuits for Pass Transistor Logic
Preprint from Proeedings of DATE 2, Paris, rane, Marh 2 On the Generation of Multiplexer Ciruits for Pass Transistor Logi Christoph Sholl Bernd Beker Institute of Computer Siene Albert Ludwigs University
More informationChapter 2: Introduction to Maple V
Chapter 2: Introdution to Maple V 2-1 Working with Maple Worksheets Try It! (p. 15) Start a Maple session with an empty worksheet. The name of the worksheet should be Untitled (1). Use one of the standard
More informationPartial Character Decoding for Improved Regular Expression Matching in FPGAs
Partial Charater Deoding for Improved Regular Expression Mathing in FPGAs Peter Sutton Shool of Information Tehnology and Eletrial Engineering The University of Queensland Brisbane, Queensland, 4072, Australia
More informationWeak Dependence on Initialization in Mixture of Linear Regressions
Proeedings of the International MultiConferene of Engineers and Computer Sientists 8 Vol I IMECS 8, Marh -6, 8, Hong Kong Weak Dependene on Initialization in Mixture of Linear Regressions Ryohei Nakano
More informationIMPROVED FUZZY CLUSTERING METHOD BASED ON INTUITIONISTIC FUZZY PARTICLE SWARM OPTIMIZATION
Journal of Theoretial and Applied Information Tehnology IMPROVED FUZZY CLUSTERING METHOD BASED ON INTUITIONISTIC FUZZY PARTICLE SWARM OPTIMIZATION V.KUMUTHA, 2 S. PALANIAMMAL D.J. Aademy For Managerial
More informationMulti-Channel Wireless Networks: Capacity and Protocols
Multi-Channel Wireless Networks: Capaity and Protools Tehnial Report April 2005 Pradeep Kyasanur Dept. of Computer Siene, and Coordinated Siene Laboratory, University of Illinois at Urbana-Champaign Email:
More informationMultiple-Criteria Decision Analysis: A Novel Rank Aggregation Method
3537 Multiple-Criteria Deision Analysis: A Novel Rank Aggregation Method Derya Yiltas-Kaplan Department of Computer Engineering, Istanbul University, 34320, Avilar, Istanbul, Turkey Email: dyiltas@ istanbul.edu.tr
More informationPerformance Benchmarks for an Interactive Video-on-Demand System
Performane Benhmarks for an Interative Video-on-Demand System. Guo,P.G.Taylor,E.W.M.Wong,S.Chan,M.Zukerman andk.s.tang ARC Speial Researh Centre for Ultra-Broadband Information Networks (CUBIN) Department
More informationA Unified Subdivision Scheme for Polygonal Modeling
EUROGRAPHICS 2 / A. Chalmers and T.-M. Rhyne (Guest Editors) Volume 2 (2), Number 3 A Unified Subdivision Sheme for Polygonal Modeling Jérôme Maillot Jos Stam Alias Wavefront Alias Wavefront 2 King St.
More informationIncremental Mining of Partial Periodic Patterns in Time-series Databases
CERIAS Teh Report 2000-03 Inremental Mining of Partial Periodi Patterns in Time-series Dataases Mohamed G. Elfeky Center for Eduation and Researh in Information Assurane and Seurity Purdue University,
More informationTime delay estimation of reverberant meeting speech: on the use of multichannel linear prediction
University of Wollongong Researh Online Faulty of Informatis - apers (Arhive) Faulty of Engineering and Information Sienes 7 Time delay estimation of reverberant meeting speeh: on the use of multihannel
More information3-D IMAGE MODELS AND COMPRESSION - SYNTHETIC HYBRID OR NATURAL FIT?
3-D IMAGE MODELS AND COMPRESSION - SYNTHETIC HYBRID OR NATURAL FIT? Bernd Girod, Peter Eisert, Marus Magnor, Ekehard Steinbah, Thomas Wiegand Te {girod eommuniations Laboratory, University of Erlangen-Nuremberg
More informationDetecting Outliers in High-Dimensional Datasets with Mixed Attributes
Deteting Outliers in High-Dimensional Datasets with Mixed Attributes A. Koufakou, M. Georgiopoulos, and G.C. Anagnostopoulos 2 Shool of EECS, University of Central Florida, Orlando, FL, USA 2 Dept. of
More informationSpace- and Time-Efficient BDD Construction via Working Set Control
Spae- and Time-Effiient BDD Constrution via Working Set Control Bwolen Yang Yirng-An Chen Randal E. Bryant David R. O Hallaron Computer Siene Department Carnegie Mellon University Pittsburgh, PA 15213.
More informationRotation Invariant Spherical Harmonic Representation of 3D Shape Descriptors
Eurographis Symposium on Geometry Proessing (003) L. Kobbelt, P. Shröder, H. Hoppe (Editors) Rotation Invariant Spherial Harmoni Representation of 3D Shape Desriptors Mihael Kazhdan, Thomas Funkhouser,
More informationCapturing Large Intra-class Variations of Biometric Data by Template Co-updating
Capturing Large Intra-lass Variations of Biometri Data by Template Co-updating Ajita Rattani University of Cagliari Piazza d'armi, Cagliari, Italy ajita.rattani@diee.unia.it Gian Lua Marialis University
More informationHIGHER ORDER full-wave three-dimensional (3-D) large-domain techniques in
FACTA UNIVERSITATIS (NIŠ) SER.: ELEC. ENERG. vol. 21, no. 2, August 2008, 209-220 Comparison of Higher Order FEM and MoM/SIE Approahes in Analyses of Closed- and Open-Region Eletromagneti Problems Milan
More information1. Introduction. 2. The Probable Stope Algorithm
1. Introdution Optimization in underground mine design has reeived less attention than that in open pit mines. This is mostly due to the diversity o underground mining methods and omplexity o underground
More informationAcoustic Links. Maximizing Channel Utilization for Underwater
Maximizing Channel Utilization for Underwater Aousti Links Albert F Hairris III Davide G. B. Meneghetti Adihele Zorzi Department of Information Engineering University of Padova, Italy Email: {harris,davide.meneghetti,zorzi}@dei.unipd.it
More informationCompilation Lecture 11a. Register Allocation Noam Rinetzky. Text book: Modern compiler implementation in C Andrew A.
Compilation 0368-3133 Leture 11a Text book: Modern ompiler implementation in C Andrew A. Appel Register Alloation Noam Rinetzky 1 Registers Dediated memory loations that an be aessed quikly, an have omputations
More informationImproved flooding of broadcast messages using extended multipoint relaying
Improved flooding of broadast messages using extended multipoint relaying Pere Montolio Aranda a, Joaquin Garia-Alfaro a,b, David Megías a a Universitat Oberta de Catalunya, Estudis d Informàtia, Mulimèdia
More informationFuzzy Meta Node Fuzzy Metagraph and its Cluster Analysis
Journal of Computer Siene 4 (): 9-97, 008 ISSN 549-3636 008 Siene Publiations Fuzzy Meta Node Fuzzy Metagraph and its Cluster Analysis Deepti Gaur, Aditya Shastri and Ranjit Biswas Department of Computer
More informationDiscrete sequential models and CRFs. 1 Case Study: Supervised Part-of-Speech Tagging
0-708: Probabilisti Graphial Models 0-708, Spring 204 Disrete sequential models and CRFs Leturer: Eri P. Xing Sribes: Pankesh Bamotra, Xuanhong Li Case Study: Supervised Part-of-Speeh Tagging The supervised
More informationDiffusion Kernels on Graphs and Other Discrete Structures
Diffusion Kernels on Graphs and Other Disrete Strutures Risi Imre Kondor ohn Lafferty Shool of Computer Siene Carnegie Mellon University Pittsburgh P 523 US KONDOR@CMUEDU LFFERTY@CSCMUEDU bstrat The appliation
More informationVertex Unfoldings of Orthogonal Polyhedra: Positive, Negative, and Inconclusive Results
CCCG 2018, Winnipeg, Canada, August 8 10, 2018 Vertex Unfoldings of Orthogonal Polyhedra: Positive, Negative, and Inonlusive Results Luis A. Garia Andres Gutierrrez Isaa Ruiz Andrew Winslow Abstrat We
More informationFast Rigid Motion Segmentation via Incrementally-Complex Local Models
Fast Rigid Motion Segmentation via Inrementally-Complex Loal Models Fernando Flores-Mangas Allan D. Jepson Department of Computer Siene, University of Toronto {mangas,jepson}@s.toronto.edu Abstrat The
More information1. The collection of the vowels in the word probability. 2. The collection of real numbers that satisfy the equation x 9 = 0.
C HPTER 1 SETS I. DEFINITION OF SET We begin our study of probability with the disussion of the basi onept of set. We assume that there is a ommon understanding of what is meant by the notion of a olletion
More informationNew Fuzzy Object Segmentation Algorithm for Video Sequences *
JOURNAL OF INFORMATION SCIENCE AND ENGINEERING 24, 521-537 (2008) New Fuzzy Obet Segmentation Algorithm for Video Sequenes * KUO-LIANG CHUNG, SHIH-WEI YU, HSUEH-JU YEH, YONG-HUAI HUANG AND TA-JEN YAO Department
More informationAlgorithms, Mechanisms and Procedures for the Computer-aided Project Generation System
Algorithms, Mehanisms and Proedures for the Computer-aided Projet Generation System Anton O. Butko 1*, Aleksandr P. Briukhovetskii 2, Dmitry E. Grigoriev 2# and Konstantin S. Kalashnikov 3 1 Department
More informationRecommendation Subgraphs for Web Discovery
Reommation Subgraphs for Web Disovery Arda Antikaioglu Department of Mathematis Carnegie Mellon University aantika@andrew.mu.edu R. Ravi Tepper Shool of Business Carnegie Mellon University ravi@mu.edu
More informationCreating Adaptive Web Sites Through Usage-Based Clustering of URLs
Creating Adaptive Web Sites Through Usage-Based Clustering of URLs Bamshad Mobasher Dept. of Computer Siene, DePaul University, Chiago, IL mobasher@s.depaul.edu Robert Cooley, Jaideep Srivastava Dept.
More informationScalable Clustering of Signed Networks Using Balance Normalized Cut
Scalable Clustering of Signed Networks Using Balance Normalized Cut Kai-Yang Chiang,, Inderjit S. Dhillon The 21st ACM International Conference on Information and Knowledge Management (CIKM 2012) Oct.
More informationWeighted Graph Cuts without Eigenvectors: A Multilevel Approach
1 Weighted Graph Cuts without Eigenvectors: A Multilevel Approach Inderjit S Dhillon, Member, IEEE, Yuqiang Guan and Brian Kulis Abstract A variety of clustering algorithms have recently been proposed
More informationParticle Swarm Optimization for the Design of High Diffraction Efficient Holographic Grating
Original Artile Partile Swarm Optimization for the Design of High Diffration Effiient Holographi Grating A.K. Tripathy 1, S.K. Das, M. Sundaray 3 and S.K. Tripathy* 4 1, Department of Computer Siene, Berhampur
More informationAn Interactive-Voting Based Map Matching Algorithm
Eleventh International Conferene on Mobile Data Management An Interative-Voting Based Map Mathing Algorithm Jing Yuan* University of Siene and Tehnology of China Hefei, China yuanjing@mail.ust.edu.n Yu
More informationCOST PERFORMANCE ASPECTS OF CCD FAST AUXILIARY MEMORY
COST PERFORMANCE ASPECTS OF CCD FAST AUXILIARY MEMORY Dileep P, Bhondarkor Texas Instruments Inorporated Dallas, Texas ABSTRACT Charge oupled devies (CCD's) hove been mentioned as potential fast auxiliary
More informationDetection and Recognition of Non-Occluded Objects using Signature Map
6th WSEAS International Conferene on CIRCUITS, SYSTEMS, ELECTRONICS,CONTROL & SIGNAL PROCESSING, Cairo, Egypt, De 9-31, 007 65 Detetion and Reognition of Non-Oluded Objets using Signature Map Sangbum Park,
More information