### Citations

14077 |
Computers and Intractability: A Guide to the Theory of NP-Completeness
- Garey, Johnson
- 1979
(Show Context)
Citation Context ...is dissertation, we are dealing with many discrete optimization problems. To gain an insight into the difficulty of those problems, performing computational complexity analysis is necessary. The book =-=[18]-=- provides a perfect guide to the computational complexity theory. Definition 3.5 P, also known as PTIME or DTIME, is one of the most fundamental complexity classes. It contains all decision problems w... |

7725 |
Matrix analysis
- Horn, Johnson
- 1985
(Show Context)
Citation Context ... matrix decomposition can be used for compressing data. The underlying reason is that if we find Am×n = Xm×k · Ck×n and k is much less than m and n, storing X and C instead of A will save great space =-=[29]-=-. 17 2.4 Nonnegative Matrix Factorization While the SVD is optimal in terms of the Frobenius norm, recently, people have realized that it does not have sufficient interpretability. To address this pro... |

5156 | Optimization by simulated annealing
- Kirkpatrick, Gelatt, et al.
- 1983
(Show Context)
Citation Context ...sed for combinatorial optimization in which an optimal solution is sought over a discrete search space. Popular heuristics for combinatorial problems include simulated annealing by Kirkpatrick et al. =-=[33]-=-, genetic algorithms by Holland et al. [28], ant colony optimization by Dorigo,[9] and tabu search by Glover [20]. 30 CHAPTER 4 BOOLEAN MATRIX DECOMPOSITION 4.1 BMD Variants with Applications 4.1.1 Ba... |

4363 | Latent Dirichlet allocation
- Blei, Ng, et al.
(Show Context)
Citation Context ... realized that it does not have sufficient interpretability. To address this problem, multiple new methods were proposed, like Probabilistic Latent Semantic Indexing [27], Latent Dirichlet Allocation =-=[9]-=- and Nonnegative Matrix Factorization (NMF) [31]. NMF is also an old problem that has been extensively studied in [55]. In NMF, the added restriction is that all the matrices should be non-negative. T... |

3884 |
Adaptation in Natural and Artificial Systems
- Holland
- 1975
(Show Context)
Citation Context ... an optimal solution is sought over a discrete search space. Popular heuristics for combinatorial problems include simulated annealing by Kirkpatrick et al. [33], genetic algorithms by Holland et al. =-=[28]-=-, ant colony optimization by Dorigo,[9] and tabu search by Glover [20]. 30 CHAPTER 4 BOOLEAN MATRIX DECOMPOSITION 4.1 BMD Variants with Applications 4.1.1 Basic BMD The basic BMD problem is to decompo... |

3778 | Indexing by latent semantic analysis
- Deerwester, Dumais, et al.
- 1990
(Show Context)
Citation Context ...ation to model user rating profiles for collaborative filtering [45, 46]. Their core idea is to find a good factorization of the input matrix to reconstruct the missing 19 cells in the input data. In =-=[8, 14, 27]-=-, they use the concept of matrix factorization to find latent factors by which to index documents. It can be applied in clustering as well. Some work even tried clustering two attributes of the input ... |

2196 |
An Introduction to Multivariate Statistical Analysis
- Anderson
- 2003
(Show Context)
Citation Context ... 2.6 Probabilistic Matrix Factorization Probabilistic matrix factorization is viewing the input matrix from the statistic perspective. The most famous work might be principle component analysis (PCA) =-=[3]-=-. It transforms a number of possibly correlated variables into a smaller number of uncorrelated variables called principal components. PCA involves the calculation of the eigenvalue decomposition of a... |

1686 |
Learning the parts of objects by non-negative matrix factorization. Nature
- Lee, Seung
- 1999
(Show Context)
Citation Context ...n describe the probabilistic relationships between individual points and centroids. Ding et al. [15] show the equivalence of NMF, spectral clustering and K-means clustering. The work of Lee and Seung =-=[38, 39]-=- also helped bring much attention from machine learning and data mining research communities to NMF. 2.5 Boolean Matrix Factorization Since many real applications involve Boolean data, such as documen... |

1518 | RoleBased Access Control Models
- Sandhu, Coyne, et al.
- 1996
(Show Context)
Citation Context ...iciency and flexibility, it has been used by the majority of enterprisers and is a newer alternative approach to mandatory access control (MAC) and discretionary access control (DAC). Sandu R. et al. =-=[57]-=- is one of the most cited papers in the field of RBAC. It defined RBAC0, the basic model, RBAC1 which introduces role hierarchies, RBAC2 which introduces constraints at the basis of RBAC0, and RBAC3 w... |

1281 |
Linear Programming and Extensions
- Dantzig
- 1998
(Show Context)
Citation Context ...it is bounded. Linear programs can be expressed in the following form. maximize(minimize) cTx (3.2) subject to Ax ≤ b (3.3) There are several good algorithms for linear program. The simplex algorithm =-=[12]-=-, developed by George Dantzig in 1947, solves LP problems by constructing a feasible solution at a vertex of the polytope and then walking along 28 a path on the edges of the polytope to vertices with... |

1243 | Algorithms for non-negative matrix factorization
- Seung, Lee
(Show Context)
Citation Context ...n describe the probabilistic relationships between individual points and centroids. Ding et al. [15] show the equivalence of NMF, spectral clustering and K-means clustering. The work of Lee and Seung =-=[38, 39]-=- also helped bring much attention from machine learning and data mining research communities to NMF. 2.5 Boolean Matrix Factorization Since many real applications involve Boolean data, such as documen... |

1225 | Probabilistic latent semantic indexing
- Hofmann
- 1999
(Show Context)
Citation Context ...benius norm, recently, people have realized that it does not have sufficient interpretability. To address this problem, multiple new methods were proposed, like Probabilistic Latent Semantic Indexing =-=[27]-=-, Latent Dirichlet Allocation [9] and Nonnegative Matrix Factorization (NMF) [31]. NMF is also an old problem that has been extensively studied in [55]. In NMF, the added restriction is that all the m... |

858 | A new polynomial-time algorithm for linear programming
- Karmarkar
- 1984
(Show Context)
Citation Context ...4.3 Linear Programming Relaxation In general the integer linear programming problem is NP-hard, while the linear programming problem is not. It has polynomial algorithms such as interior point method =-=[32]-=- and also has simplex method algorithms which have very good practical performance despite exponential worst-case running time [12]. Thus a typical approach to solve an integer linear programming prob... |

481 | Biclustering algorithms for biological data analysis: A survey
- Madeira, Oliveira
- 2004
(Show Context)
Citation Context ... to index documents. It can be applied in clustering as well. Some work even tried clustering two attributes of the input data simultaneously, by factoring a matrix into the product of three matrices =-=[17, 44, 51]-=-. 20 CHAPTER 3 BACKGROUND 3.1 Access Control In computer system security, role-based access control (RBAC) is an approach to restricting system access to authorized users. Due to its advantages in adm... |

312 | An Automatic Method of Solving Discrete Programming Problems,"
- Land, Doig
- 1960
(Show Context)
Citation Context ...r programming. First, optimization has been studied for more than half century. There are quite a few good exact optimization algorithms, even for integer linear programming, such as branch-and-bound =-=[37]-=-. In addition, successful optimization software Packages are easily obtainable, such as Matlab and the Neos server 1. Even though those BMD variants are proven to be hard to solve, small or medium siz... |

159 | On the equivalence of nonnegative matrix factorization and spectral clustering. In:
- Ding, He, et al.
- 2005
(Show Context)
Citation Context ...tion is that all the matrices should be non-negative. This can help cluster data, find centroids and even describe the probabilistic relationships between individual points and centroids. Ding et al. =-=[15]-=- show the equivalence of NMF, spectral clustering and K-means clustering. The work of Lee and Seung [38, 39] also helped bring much attention from machine learning and data mining research communities... |

151 | A systematic comparison and evaluation of biclustering methods for gene expression data.
- Prelic, Bleuer, et al.
- 2006
(Show Context)
Citation Context ...sed, like Probabilistic Latent Semantic Indexing [27], Latent Dirichlet Allocation [9] and Nonnegative Matrix Factorization (NMF) [31]. NMF is also an old problem that has been extensively studied in =-=[55]-=-. In NMF, the added restriction is that all the matrices should be non-negative. This can help cluster data, find centroids and even describe the probabilistic relationships between individual points ... |

142 |
Approximation Algorithms for the Set Covering and Vertex Cover Problems
- Hochbaum
- 1982
(Show Context)
Citation Context ...ved through traditional optimization techniques. In addition to exact algorithms, approximation algorithms may be developed through LP-based techniques, such as dual-fitting [41], randomized rounding =-=[23]-=-, and primal-dual schema [36]. The linear programming framework we will propose in fact can not only incorporate BMD variants, but also problems in other application domains, such as tiling database p... |

116 | The maximum edge biclique problem is np-complete.
- Peeters
- 2003
(Show Context)
Citation Context ...an be related to many research problems that have been studied in the literature. Those problems can be viewed as either its special case or its variant. Definition 6.5 (Maximum Edge Biclique Problem =-=[53]-=-) Given a bipartite graph G = (V1∪V2, E) and a positive integer K, does G contain a biclique with at least k edges? 114 A bicilique is a kind of bipartite graph where every vertex of the first set is ... |

104 | Modeling user rating profiles for collaborative filtering.
- Marlin
- 2003
(Show Context)
Citation Context ...composition of a data covariance matrix or singular value decomposition of a data matrix. Some work were proposed to use matrix factorization to model user rating profiles for collaborative filtering =-=[45, 46]-=-. Their core idea is to find a good factorization of the input matrix to reconstruct the missing 19 cells in the input data. In [8, 14, 27], they use the concept of matrix factorization to find latent... |

82 | Text mining: The state of the art and the challenges.
- Tan
(Show Context)
Citation Context ...xt clustering, concept/entity extraction, production of granular taxonomies, sentiment analysis, document summarization, and entity relation modeling (i.e., learning relations between named entities) =-=[30]-=-. For in16 stance, data summarization is becoming an very important research topic, because with the widespread of internet technology we are facing the dramatically increasing amount of document data... |

68 |
Simple and fast algorithms for linear and integer programs with two variables per inequality.
- Hochbaum, Naor
- 1994
(Show Context)
Citation Context ...ves a linear programming formulation for the edge weighted biclique problem which is to delete from a bipartite graph {V1, V2, E}, 1For the definition of a monotone integer system, refer to the paper =-=[24]-=-. 124 a minimum weight collection of edges so that the remaining edges induce a complete bipartite graph. By slight modifications, we obtain a linear programming formulation for our special case as be... |

64 |
On the ratio of optimal integral and fractional covers. Discrete mathematics
- Lovász
- 1975
(Show Context)
Citation Context ... problems can still be solved through traditional optimization techniques. In addition to exact algorithms, approximation algorithms may be developed through LP-based techniques, such as dual-fitting =-=[41]-=-, randomized rounding [23], and primal-dual schema [36]. The linear programming framework we will propose in fact can not only incorporate BMD variants, but also problems in other application domains,... |

63 | The role-based access control system of a european bank: A case study and discussion.
- Schaad, Moffett, et al.
- 2001
(Show Context)
Citation Context ...atic work for determining roles. Conceptually, there are two types of approaches towards role engineering. They are top-down and bottom-up. Top-down is by analyzing business processes to deduce roles =-=[1, 2, 11]-=-. However, all those approaches share one common weakness that it ignores existing user-to-permission assignments and calls for the cooperation among various authorities from different disciplines. Th... |

63 | The role mining problem: Finding a minimal descriptive set of roles. - Vaidya, Atluri, et al. - 2007 |

52 | Roleminer: mining roles using subset enumeration.
- Vaidya, Atluri, et al.
- 2006
(Show Context)
Citation Context ...organizations need to define a good set of roles, and then assign them to users appropriately, the work of which is called role engineering. To automate the process of role engineering, Vaidya et al. =-=[62]-=- propose to mine roles from existing user-to-permission assignment data, which is the origin of RMP. Look at the example of user-to-permission assignment data represented by a bipartite graph as shown... |

51 | Role mining - revealing business roles for security administration using data mining technology,” in
- Kuhlmann, Shohat, et al.
- 2003
(Show Context)
Citation Context ...proach generates roles purely from the existing user-to-permission assignments. It allows the automation of role generation without knowing the semantics of busi15 ness. Kuhlmann, Shohat, and Schmipf =-=[47]-=- present a bottom-up approach using clustering technique similar to the well known k-means clustering. Schlegelmilch and Steffens [58] have proposed an agglomerative clustering based role mining appro... |

49 | On the red-blue set cover problem.
- Carr, Doddi, et al.
- 2000
(Show Context)
Citation Context ...us now look at the conservative partial SRM I problem. It is a NP-hard problem as well. For ease of understanding, we would to like to study it in the setting of the red-blue set cover problem (RBSC) =-=[10]-=-. Problem 5.7 (RBSC ) Given a finite set of red elements R and a finite set of blue elements B and a family S = {S1, ..., Sn} ∈ 2R∪B , find a subfamily C ∈ S which covers all blue elements, but which ... |

49 |
Role engineering,”
- Coyne
- 1996
(Show Context)
Citation Context ...mplete and correct set of roles. According to a study by NIST, this task has been identified as the costliest component in realizing RBAC. The concept of role engineering was first presented by Coyne =-=[11]-=-. It refers to the systematic work for determining roles. Conceptually, there are two types of approaches towards role engineering. They are top-down and bottom-up. Top-down is by analyzing business p... |

49 | Tiling databases,
- Geerts, Goethals, et al.
- 2004
(Show Context)
Citation Context ...that despite in different domains, many problems are equivalent from the perspective of BMD. For instance, the basic RMP problem in role mining [60] and the minimum tiling problem in tiling databases =-=[19]-=- are the same. Unfortunately, these problems used to be studied in their own disciplinary. So to fully exploit the benefits of BMD and help people recognize its importance, we need a general framework... |

48 |
A general model for clustering binary data.
- Li
- 2005
(Show Context)
Citation Context ...a (users vs websites), DNA microarray expression profiles and protein-protein complex interaction network [63], Boolean data have obtained a special and important space in the domain of data analysis =-=[40]-=-. It is natural to represent Boolean data by Boolean matrices, which are a special case of non-negative matrices. Many research problems involved in Boolean data analysis can be reduced to Boolean mat... |

48 | Role mining with ORCA. In
- Schlegelmilch, Steffens
- 2005
(Show Context)
Citation Context ...owing the semantics of busi15 ness. Kuhlmann, Shohat, and Schmipf [47] present a bottom-up approach using clustering technique similar to the well known k-means clustering. Schlegelmilch and Steffens =-=[58]-=- have proposed an agglomerative clustering based role mining approach, known as ORCA. More recently, Vaidya et al. [62] proposed an approach based on subset enumeration, called RoleMiner. This approac... |

45 | Tight bounds and 2-approximation algorithms for integer programs with two variables per inequality
- Hochbaum, Megiddo, et al.
- 1993
(Show Context)
Citation Context ...ject to m linear constraints with at most two variables per inequality, and with all variables bounded between 0 and U , has a 2-approximation algorithm which runs in polynomial time O(mnU2log(Un2m)) =-=[26]-=-. Hochaum et al. in [26] present a 2-approximation algorithm for integer linear programs with at most two variables per inequality. We briefly introduce this algorithm here. It transforms the integer ... |

44 | Approximating clique and biclique problems,
- Hochbaum
- 1998
(Show Context)
Citation Context ...alue of 1 in a row of A. As we have shown in the preceding section, the dominant discrete pattern mining problem with w1 = 1 and w2 ≥ T is equivalent to the maximum edge biclique problem. Hochbaun in =-=[25]-=- gives a linear programming formulation for the edge weighted biclique problem which is to delete from a bipartite graph {V1, V2, E}, 1For the definition of a monotone integer system, refer to the pap... |

43 |
Adaptive memory tabu search for binary quadratic programs
- Glover, Kochenberger, et al.
- 1998
(Show Context)
Citation Context ...iscrete pattern mining problem can be formulated as an unconstrained quadratic binary program. In literature, tabu search has been successfully employed to solve quadratic binary programming problems =-=[21]-=-. This is another reason why we choose tabu search. The dominant discrete pattern mining problem is given a binary matrix A to maximize ∑ i S(Ai:, Y ). It is easy to compute S(Ai:, Y ) once Y is deter... |

42 |
Survey of Text Mining
- Berry
- 2003
(Show Context)
Citation Context ...here each topic could simply be a subset of words or a representative document. A good summarization scheme not only reduces data storage space, but also facilitates the task of information retrieval =-=[5]-=-. 2.3 Ordinary Matrix Factorization Ordinary matrix decomposition is a well-studied problem that has been the focus of significant research. Indeed, one of the earliest motivations of matrix decomposi... |

41 | The discrete basis problem,
- Miettinen, Mielikainen, et al.
(Show Context)
Citation Context ... programming framework we will propose in fact can not only incorporate BMD variants, but also problems in other application domains, such as tiling database problems [19] and discrete basis problems =-=[50]-=-. 1http://www-neos.mcs.anl.gov/ 42 For ease of explaining, in this section, we will discuss BMD variants in the role mining context, where a BMD solution {X,C} of user-to-permission assignments A give... |

38 |
The multiple multiplicative factor model for collaborative filtering.
- Marlin, Zemel
- 2004
(Show Context)
Citation Context ...composition of a data covariance matrix or singular value decomposition of a data matrix. Some work were proposed to use matrix factorization to model user rating profiles for collaborative filtering =-=[45, 46]-=-. Their core idea is to find a good factorization of the input matrix to reconstruct the missing 19 cells in the input data. In [8, 14, 27], they use the concept of matrix factorization to find latent... |

31 | An Extended Authorization Model for Relational Databases.
- Bertino, Samarati, et al.
- 1997
(Show Context)
Citation Context ...include both positive and negative assignments, as semantic user-role assignments. Indeed, negative authorizations are integral part of many access control sys70 tems. From the work of Bertino et al. =-=[6, 7]-=-, introducing negative authorizations have many advantages. They enable a temporary suspension of a permission from a user without having to revoke it (revoking a permission sometimes may have a casca... |

29 | Fast exact and heuristic methods for role minimization problems.
- Ene, Horne, et al.
- 2008
(Show Context)
Citation Context ... results are illustrated as shown in Figure 4.6. All four graphs suggest that Two-Stage is better than Edge-Key. 4.5.2 Real Data We run our greedy heuristics on real data sets collected by Ene et al. =-=[16]-=-. They are emea, healthcare, domino, firewall 1, and firewall 2. The first experiment is to study the basic BMD. We run Algorithm 4.3 against each data set with candidate roles generated by Itself and... |

26 | Optimal boolean matrix decomposition. application to role engineering,” in
- Lu, Vaidya, et al.
- 2008
(Show Context)
Citation Context ...red roles are most advantageous to implement. To find the optimal role set matching the interestingness measures , Vaidya et al. [60] took a step forward to propose the role mining problem. Lu et al. =-=[42]-=- further connect the role mining problem with the Boolean matrix decomposition problem. 2.2 Text Mining Text mining, roughly equivalent to text analytic, refers generally to the process of deriving hi... |

22 |
Authorizations in relational database management systems.
- Bertino, Samarati, et al.
- 1993
(Show Context)
Citation Context ...include both positive and negative assignments, as semantic user-role assignments. Indeed, negative authorizations are integral part of many access control sys70 tems. From the work of Bertino et al. =-=[6, 7]-=-, introducing negative authorizations have many advantages. They enable a temporary suspension of a permission from a user without having to revoke it (revoking a permission sometimes may have a casca... |

14 | J.M.: A class of probabilistic models for role engineering. In:
- Frank, Basin, et al.
- 2008
(Show Context)
Citation Context ... to index documents. It can be applied in clustering as well. Some work even tried clustering two attributes of the input data simultaneously, by factoring a matrix into the product of three matrices =-=[17, 44, 51]-=-. 20 CHAPTER 3 BACKGROUND 3.1 Access Control In computer system security, role-based access control (RBAC) is an approach to restricting system access to authorized users. Due to its advantages in adm... |

14 | Binary matrix factorization with applications,”
- Zhang, Li, et al.
- 2007
(Show Context)
Citation Context ...ny real applications involve Boolean data, such as document-to-term data, web click-stream data (users vs websites), DNA microarray expression profiles and protein-protein complex interaction network =-=[63]-=-, Boolean data have obtained a special and important space in the domain of data analysis [40]. It is natural to represent Boolean data by Boolean matrices, which are a special case of non-negative ma... |

13 | On finding large conjunctive clusters
- Mishra, Ron, et al.
(Show Context)
Citation Context ... to index documents. It can be applied in clustering as well. Some work even tried clustering two attributes of the input data simultaneously, by factoring a matrix into the product of three matrices =-=[17, 44, 51]-=-. 20 CHAPTER 3 BACKGROUND 3.1 Access Control In computer system security, role-based access control (RBAC) is an approach to restricting system access to authorized users. Due to its advantages in adm... |

10 | Topics in 0–1 data,
- Bingham, Mannila, et al.
- 2002
(Show Context)
Citation Context ...ation to model user rating profiles for collaborative filtering [45, 46]. Their core idea is to find a good factorization of the input matrix to reconstruct the missing 19 cells in the input data. In =-=[8, 14, 27]-=-, they use the concept of matrix factorization to find latent factors by which to index documents. It can be applied in clustering as well. Some work even tried clustering two attributes of the input ... |

10 | On the positive–negative partial set cover problem - MIETTINEN |

8 | Interpretable nonnegative matrix decompositions
- Hyvönen, Miettinen, et al.
- 2008
(Show Context)
Citation Context ...rpretability. To address this problem, multiple new methods were proposed, like Probabilistic Latent Semantic Indexing [27], Latent Dirichlet Allocation [9] and Nonnegative Matrix Factorization (NMF) =-=[31]-=-. NMF is also an old problem that has been extensively studied in [55]. In NMF, the added restriction is that all the matrices should be non-negative. This can help cluster data, find centroids and ev... |

8 | Compression, clustering, and pattern discovery in very digh-dimensional discrete-attribute data sets
- Koyutürk, Grama, et al.
- 2005
(Show Context)
Citation Context ...tterns are identified. The task of analyzing the original binary matrix can be switched on those mined patterns, which have much smaller size. This technique has received increased attention recently =-=[35, 59]-=-. For example, Shen et al. [59] even provide an efficient 2-approximation algorithm for rank-one binary matrix approximation recursively conducted in a process of PROXIMUS. Rank-one binary matrix appr... |

8 |
The Boolean column and column-row matrix decompositions,
- Miettinen
- 2008
(Show Context)
Citation Context ...f 1. Miettinen et al. [52] consider C as a discrete basis of A, from which A can be reconstructed. Given A, how to find a good basis is their core problem. This work is further developed by Miettinen =-=[48]-=- where C is limited to be the subset of columns of A. This limitation gives increased interpretability since each column of C can be seen as a centroid of A from the perspective of clustering. [48] al... |

6 | PROXIMUS: a framework for analyzing very high dimensional discrete-attributed datasets.
- Koyuturk, Grama
- 2003
(Show Context)
Citation Context ...Frobenius norm of the difference. However, result interpretation is difficult for datasets with discrete attributes. Existing techniques for mining discrete patterns from binary data include PROXIMUS =-=[34]-=-, which can serve the purpose of data compression as well. The idea is to decomposes a given binary matrix into a pattern vector and a presence vector, which are restricted to be binary, such that the... |

6 |
On the biclique problem in bipartite graphs, GSIA Working Paper 1996-04,
- Dawande, Keskinocak, et al.
- 1996
(Show Context)
Citation Context ...largest tile induced by the dominant discrete pattern is as shown in Equation (6.7). 1 0 1 1 0 1 1 1 1 1 0 0 1 0 0 1 1 1 1 0 (6.7) Definition 6.7 (Maximum Edge Weight Biclique Problem =-=[43]-=-) Given a bipartite graph {V1 ∪ V2, E} and weights {wij} associated with edges {(V1(i), V2(j))} respectively, find a biclique C, where the sum of the edge weights is maximum. 116 The dominant discrete... |

6 | Mining discrete patterns via binary matrix factorization
- Shen, Ji, et al.
- 2009
(Show Context)
Citation Context ...tterns are identified. The task of analyzing the original binary matrix can be switched on those mined patterns, which have much smaller size. This technique has received increased attention recently =-=[35, 59]-=-. For example, Shen et al. [59] even provide an efficient 2-approximation algorithm for rank-one binary matrix approximation recursively conducted in a process of PROXIMUS. Rank-one binary matrix appr... |

4 | Solving the maximum edge weight clique problem via unconstrained quadratic programming
- Alidaee, Glover, et al.
(Show Context)
Citation Context ...ary matrix is closely related to a bipartite graph. It is also a useful approach for problems involved with a bipartite graph to be formulated as an unconstrained quadratic binary programming problem =-=[4]-=-. 6.4.2 Integer Linear Programming The unconstrained quadratic binary programming formulation can be linearized as an integer linear programming problem by introducing auxiliary binary variables {Zij}... |

2 |
Observations on the role lifecyle in the context of enterprise security managemen
- KERN, KUHLMANN, et al.
- 2002
(Show Context)
Citation Context ...atic work for determining roles. Conceptually, there are two types of approaches towards role engineering. They are top-down and bottom-up. Top-down is by analyzing business processes to deduce roles =-=[1, 2, 11]-=-. However, all those approaches share one common weakness that it ignores existing user-to-permission assignments and calls for the cooperation among various authorities from different disciplines. Th... |

2 |
Tabu search ł part i
- GLOVER
- 1989
(Show Context)
Citation Context ...euristics for combinatorial problems include simulated annealing by Kirkpatrick et al. [33], genetic algorithms by Holland et al. [28], ant colony optimization by Dorigo,[9] and tabu search by Glover =-=[20]-=-. 30 CHAPTER 4 BOOLEAN MATRIX DECOMPOSITION 4.1 BMD Variants with Applications 4.1.1 Basic BMD The basic BMD problem is to decompose an input Boolean matrix into two Boolean matrices with the minimum ... |

2 | Bayesian computation and stochastic systems
- PETER, GREEN, et al.
- 1995
(Show Context)
Citation Context ...V (x)} exp{Clog(1+n)V (y)} }; 8: n← n+ 1; 9: end while value of λ to change with time. Simulated annealing is a popular variation of the preceding. Here, we adopt the formula proposed by Besag et al. =-=[54]-=- and let the transition probability be min{1, exp{λnV (x)/N(x)} exp{λnV (y)/N(y)} } where λn = log(1 + n). In our case, N(x) and N(y) are equivalent and are canceled out in the formula. As computing t... |

2 |
Edge-rmp: Minimizing administrative assignments for role-based access control.
- Vaidya, Atluri, et al.
- 2009
(Show Context)
Citation Context ...imize the administrative cost ||X||1+||C||1, while basic BMD only aims to minimize the number of roles. So a basic BMD optimal solution is not necessarily a cost BMD solution. Cost BMD was studied in =-=[61]-=-, where a greedy heuristic was proposed. To distinguish it, we call it Edge-Key. Here we propose a new heuristic, called Two-Stage. 55 Cost BMD objective consists of two parts ||X||1 and ||C||1. Intui... |

1 |
Matrix Computations. The Johns Hopkins Universtiy
- GOLUB, LOAN
- 1996
(Show Context)
Citation Context ...} of variables X and accordingly the constraint set can be reformed as BXB + NXN = b. By letting XN = 0, XB = B−1b. If B−1b > 0, {B−1b, 0} is a basic feasible solution. According to the Cramer’s rule =-=[22]-=-, the ith competent of XB is det(B ′) det(B) , where det(B) denotes the determinant of B, B′ is B except its ith column is replaced by b. Because in Equation (6.11) the coefficient matrix of the const... |

1 |
The hungarian method for the assgnment problem. Navel Research Logistics Quarterly
- KUHN
- 1955
(Show Context)
Citation Context ...ization techniques. In addition to exact algorithms, approximation algorithms may be developed through LP-based techniques, such as dual-fitting [41], randomized rounding [23], and primal-dual schema =-=[36]-=-. The linear programming framework we will propose in fact can not only incorporate BMD variants, but also problems in other application domains, such as tiling database problems [19] and discrete bas... |

1 |
The discrete basis problem
- PAULI, TANELI, et al.
(Show Context)
Citation Context ...g databases problem is reduced to finding a factorization of A = C ⊗ X by limiting each column of 18 C to be a subset of one column of A, because each tile can only cover cells of 1. Miettinen et al. =-=[52]-=- consider C as a discrete basis of A, from which A can be reconstructed. Given A, how to find a good basis is their core problem. This work is further developed by Miettinen [48] where C is limited to... |

1 |
Third Edition (Statistical Modeling and Decision Science) (Hardcover
- ROSS
- 2002
(Show Context)
Citation Context ...onstruction error with the solution t, and N(t) is the number of neighboring values of t. Such a Markov Chain has a limiting probability of 1 for arriving at optimal minimization solutions when λ → ∞ =-=[56]-=-. But it has been found to be more useful or efficient to allow the 52 Algorithm 4.2 Simulated Annealing Algorithm for Partial BMD Input: A1×n and Cr×n Output: X1×r 1: X, x← (0, ..., 0); n← 1; 2: whil... |