#### DMCA

## Speeding Up Permutation Based Indexing with Indexing

Citations: | 3 - 0 self |

### Citations

493 |
The FERET database and evaluation procedure for face recognition algorithms
- Phillips, Wechsler, et al.
- 1998
(Show Context)
Citation Context ...milar permutations is constant (as the function of k). Even so, the slow-down is quite mild for the k values shown. We experimented with a real database using Euclidean distance. This database (FERET =-=[19]-=- 2 ) is composed of vectors of features extracted from real (face) images. The vectors have 761 components. The database size was a set of 762 faces, plus another (distinct) set of 254 faces used for ... |

429 | Searching in metric spaces - Chávez, Navarro, et al. |

352 | Data structures and algorithms for nearest neighbor search in general metric spaces
- Yianilos
- 1993
(Show Context)
Citation Context ...ane Tree [7]. The original method used binary trees, while the implementation we used is generalized to arbitrary arity; we used the same arity as the dimension of the space. Multi Vantage-Point tree =-=[17]-=-, [6]. We used bucket size 12 and the tree arity was the same as the metric space dimension was. Spatial Approximation Tree [9].lcluster List of Clusters [18], with bucket size 12. Unless otherwise s... |

265 | Comparing top k lists
- Fagin, Kumar, et al.
- 2003
(Show Context)
Citation Context ...mpared against the query in this order, until some stopping criterion is met. Similarity between two permutations can be measured for example by Kendall Tau, Spearman Rho, or Spearman Footrule metric =-=[15]-=-. All of them are metric. We use Spearman Rho because it is not expensive to compute and according to the authors in [1] it has a good performance to predict proximity between elements. (The square of... |

214 | Near neighbor search in large metric spaces
- Brin
- 1995
(Show Context)
Citation Context ... defined to belong to the group i if pi is the closest center. Again, given that pi is the closest center to q, we can discard the whole group j if d(q, pj) − r > d(q, pi) + r. Some examples are [7], =-=[8]-=-, [9]. C. Approximate and Probabilistic Algorithms In approximate algorithms one usually has a parameter ε, so that the retrieved elements are guaranteed to have a distance to the query at most (1 + ε... |

204 |
Satisfying general proximity/similarity queries with metric trees. Information processing letters
- Uhlmann
- 1991
(Show Context)
Citation Context ...an be defined to belong to the group i if pi is the closest center. Again, given that pi is the closest center to q, we can discard the whole group j if d(q, pj) − r > d(q, pi) + r. Some examples are =-=[7]-=-, [8], [9]. C. Approximate and Probabilistic Algorithms In approximate algorithms one usually has a parameter ε, so that the retrieved elements are guaranteed to have a distance to the query at most (... |

138 |
Some approaches to best-match file searching
- Burkhard, Keller
- 1973
(Show Context)
Citation Context ...equality filter out every database element u that satisfies |d(q, pi) − d(u, pi)| > r. Many variations of this basic scheme exist, having different space and time trade-offs. Some representatives are =-=[3]-=-, [4], [5], [6]. B. Compact Partition Based Algorithms In this family the space is divided into compact zones (as small as possible). A set of objects (centers) {p1, . . . , pk} ⊆ U are chosen and the... |

59 | Proximity matching using fixed-queries trees
- Baeza-Yates, Cunto, et al.
- 1994
(Show Context)
Citation Context ...ity filter out every database element u that satisfies |d(q, pi) − d(u, pi)| > r. Many variations of this basic scheme exist, having different space and time trade-offs. Some representatives are [3], =-=[4]-=-, [5], [6]. B. Compact Partition Based Algorithms In this family the space is divided into compact zones (as small as possible). A set of objects (centers) {p1, . . . , pk} ⊆ U are chosen and the rest... |

57 |
An algorithm for finding nearest neighbors in (approximately) constant average time
- Vidal
- 1986
(Show Context)
Citation Context ...ilter out every database element u that satisfies |d(q, pi) − d(u, pi)| > r. Many variations of this basic scheme exist, having different space and time trade-offs. Some representatives are [3], [4], =-=[5]-=-, [6]. B. Compact Partition Based Algorithms In this family the space is divided into compact zones (as small as possible). A set of objects (centers) {p1, . . . , pk} ⊆ U are chosen and the rest are ... |

47 | Excluded middle vantage point forests for nearest neighbor search
- Yianilos
- 1998
(Show Context)
Citation Context ... out every database element u that satisfies |d(q, pi) − d(u, pi)| > r. Many variations of this basic scheme exist, having different space and time trade-offs. Some representatives are [3], [4], [5], =-=[6]-=-. B. Compact Partition Based Algorithms In this family the space is divided into compact zones (as small as possible). A set of objects (centers) {p1, . . . , pk} ⊆ U are chosen and the rest are distr... |

41 | A compact space decomposition for effective metric indexing
- Chávez, Navarro
- 2005
(Show Context)
Citation Context ...n of the space. Multi Vantage-Point tree [17], [6]. We used bucket size 12 and the tree arity was the same as the metric space dimension was. Spatial Approximation Tree [9].lcluster List of Clusters =-=[18]-=-, with bucket size 12. Unless otherwise stated, we used the same amount of permutants as was the dimension of the real space; for example, in dimension 8 we used 8 permutants, in dimension 12 we used ... |

33 | Effective Proximity Retrieval by Ordering Permutations
- Gonzalez, Figueroa, et al.
(Show Context)
Citation Context ...2 GHz CPU and 4GB, 667 MHz of RAM with Mac OS X, version 10.5.5. A. Setting and Rationale As the performance and the accuracy of the base-line method Sequential-Query algorithm is well established in =-=[16]-=-, and our method computes exactly the same results (assuming we use an exact method for the second index), we do not compare it against the competitors. Rather, we consentrate on our improvement, i.e.... |

21 | A probabilistic spell for the curse of dimensionality
- CHÁVEZ, NAVARRO
(Show Context)
Citation Context ...Probabilistic algorithms on the other hand state that the answer is correct with high probability. Some examples are [12], [13]. It is also possible to turn any exact algorithm into probabilistic one =-=[14]-=-. A recent probabilistic method is based on sorting the database in a particular order, and then traversing only a small fraction of the database in that order [1]. We cover this in detail in the next... |

18 | Probabilistic proximity searching algorithms based on compact partitions
- Bustos, Navarro
(Show Context)
Citation Context ... has a parameter ε, so that the retrieved elements are guaranteed to have a distance to the query at most (1 + ε) times of what was asked for. This relaxation gives faster algorithms when ε increases =-=[10]-=-, [11]. Probabilistic algorithms on the other hand state that the answer is correct with high probability. Some examples are [12], [13]. It is also possible to turn any exact algorithm into probabilis... |

12 | Proximity searching in high dimensional spaces with a proximity preserving order
- Chávez, Figueroa, et al.
- 2005
(Show Context)
Citation Context ...me that are not relevant. Fortunately, the answer is usually good enough for many applications. The common aspect of these algorithms is that they usually trade time for the quality of the answer. In =-=[1]-=- a novel probabilistic technique was presented for proximity searching in metric spaces. This method has extremely good performance in high dimensional spaces. However, it makes sequential scan in a c... |

12 | t-Spanners as a data structure for metric space searching
- Navarro, Paredes, et al.
- 2002
(Show Context)
Citation Context ...he distance function is otherwise costly to evaluate. Also, for large number of permutants, one could trade space for time, by using e.g. AESA [5] or some more space efficient variants of it, such as =-=[20]-=-, [21]. Another line of work is to apply some approximate or probabilistic algorithm for the second index. We ran some preliminary experiments with using the permutations based algorithm again. Howeve... |

11 | A modification of the LAESA algorithm for approximated k-NN classification
- Moreno-Seco, Mico, et al.
(Show Context)
Citation Context ...ed for. This relaxation gives faster algorithms when ε increases [10], [11]. Probabilistic algorithms on the other hand state that the answer is correct with high probability. Some examples are [12], =-=[13]-=-. It is also possible to turn any exact algorithm into probabilistic one [14]. A recent probabilistic method is based on sorting the database in a particular order, and then traversing only a small fr... |

3 | Improvements of TLAESA nearest neighbour search algorithm and extension to approximation search
- Tokoro, Yamaguchi, et al.
- 2006
(Show Context)
Citation Context ... parameter ε, so that the retrieved elements are guaranteed to have a distance to the query at most (1 + ε) times of what was asked for. This relaxation gives faster algorithms when ε increases [10], =-=[11]-=-. Probabilistic algorithms on the other hand state that the answer is correct with high probability. Some examples are [12], [13]. It is also possible to turn any exact algorithm into probabilistic on... |

2 | Simple space-time trade-offs for aesa
- Figueroa, Fredriksson
- 2007
(Show Context)
Citation Context ...tance function is otherwise costly to evaluate. Also, for large number of permutants, one could trade space for time, by using e.g. AESA [5] or some more space efficient variants of it, such as [20], =-=[21]-=-. Another line of work is to apply some approximate or probabilistic algorithm for the second index. We ran some preliminary experiments with using the permutations based algorithm again. However, in ... |

1 |
Searching in metric spaces by spatial approximation,” The Very Large Databases
- unknown authors
- 2002
(Show Context)
Citation Context ...ned to belong to the group i if pi is the closest center. Again, given that pi is the closest center to q, we can discard the whole group j if d(q, pj) − r > d(q, pi) + r. Some examples are [7], [8], =-=[9]-=-. C. Approximate and Probabilistic Algorithms In approximate algorithms one usually has a parameter ε, so that the retrieved elements are guaranteed to have a distance to the query at most (1 + ε) tim... |