## Parallel VA-File (1997)

Citations: | 2 - 2 self |

### BibTeX

@TECHREPORT{Weber97parallelva-file,

author = {Roger Weber},

title = {Parallel VA-File},

institution = {},

year = {1997}

}

### OpenURL

### Abstract

Similarity search is one of the typical query type for multimedia retrieval, data mining and decision support systems. Many similarity measures transform objects into points in a high-dimensional vector space and define similarity of two objects with respect to their distance in the vector space. Data-partitioning index methods for such spaces like R-tree or X-tree are known to deteriorate with the increased number of dimensions. A newer approach, the so called VA-File, overcomes the difficulties of high dimensionality. However, the VA-File search method is linear in the number of objects. In this paper, we investigate parallelization issues of that linear process either by using multiple disks assigned to a single processor, or by using a cluster of workstation. Performance measurements of parallel VA-File in a multiple disk environment are very promising and show almost linear speed-up with the increased number of disks. Distributed VA-File search in a cluster of workstation not only...

### Citations

2219 | R-Trees: A Dynamic Index Structure for Spatial Searching
- Guttman
- 1984
(Show Context)
Citation Context ...samt fur Bildung und Wissenshaft (BBW, grant no. 93.0135). 1 For high-dimensional vector spaces, the conventional approach is to use a multidimensional access method such as grid-file [NHS84], R-tree =-=[Gut84]-=- and their variants [BKSS90, SRF87], X-tree [BKK96], SR-tree [KS97], TV-tree [LJF94] or M-tree [CPZ97]. In general terms, these methods partition the data space, cluster the data according to the part... |

1174 | The Design and Analysis of Spatial Data Structures - Samet - 1990 |

980 | The R*-tree: An efficient and robust access method for points and rectangles
- Beckmann, Kriegel, et al.
- 1990
(Show Context)
Citation Context ...ace during search so that the access cost becomes logarithmic in the number of objects in the tree. Some of these methods use the vectors directly to cluster objects together (R-tree [Gut84], R*-tree =-=[BKSS90]-=-, Xtree [BKK96], SR-tree [KS97], TV-tree [LJF94]) whereas other methods only use distances between objects (M-tree [CPZ97], vp-tree [Chi94]). The shape of a page can be either a hyper cube (R-tree, R*... |

514 | M-tree: An efficient Access method for Similarity Search in Metric Spaces
- Ciaccia, patella, et al.
- 1997
(Show Context)
Citation Context ...onventional approach is to use a multidimensional access method such as grid-file [NHS84], R-tree [Gut84] and their variants [BKSS90, SRF87], X-tree [BKK96], SR-tree [KS97], TV-tree [LJF94] or M-tree =-=[CPZ97]-=-. In general terms, these methods partition the data space, cluster the data according to the partitioning and use a hierarchical structure to prune the search space during querying. Unfortunately, wh... |

507 | The X-tree : An index structure for high-dimensional data
- Berchtold, Keim, et al.
- 1996
(Show Context)
Citation Context ....0135). 1 For high-dimensional vector spaces, the conventional approach is to use a multidimensional access method such as grid-file [NHS84], R-tree [Gut84] and their variants [BKSS90, SRF87], X-tree =-=[BKK96]-=-, SR-tree [KS97], TV-tree [LJF94] or M-tree [CPZ97]. In general terms, these methods partition the data space, cluster the data according to the partitioning and use a hierarchical structure to prune ... |

383 | The grid file: an adaptable, symmetric multikey file structure
- Nievergelt, Hinterberger, et al.
- 1984
(Show Context)
Citation Context ... the Swiss Bundesamt fur Bildung und Wissenshaft (BBW, grant no. 93.0135). 1 For high-dimensional vector spaces, the conventional approach is to use a multidimensional access method such as grid-file =-=[NHS84]-=-, R-tree [Gut84] and their variants [BKSS90, SRF87], X-tree [BKK96], SR-tree [KS97], TV-tree [LJF94] or M-tree [CPZ97]. In general terms, these methods partition the data space, cluster the data accor... |

375 | SR-tree: an index structure for high-dimensional nearest neighbour queries
- Katayama, Satoh
- 1997
(Show Context)
Citation Context ...h-dimensional vector spaces, the conventional approach is to use a multidimensional access method such as grid-file [NHS84], R-tree [Gut84] and their variants [BKSS90, SRF87], X-tree [BKK96], SR-tree =-=[KS97]-=-, TV-tree [LJF94] or M-tree [CPZ97]. In general terms, these methods partition the data space, cluster the data according to the partitioning and use a hierarchical structure to prune the search space... |

363 | Similarity of color images - Stricker, Orengo - 1995 |

262 | The R+-Tree: A Dynamic Index for Multi-Dimensional Objects," VLDB - Sellis, Roussopoulos, et al. - 1987 |

195 | The tv-tree: An index structure for high-dimensional data
- Lin, Jagadish, et al.
- 1994
(Show Context)
Citation Context ...ctor spaces, the conventional approach is to use a multidimensional access method such as grid-file [NHS84], R-tree [Gut84] and their variants [BKSS90, SRF87], X-tree [BKK96], SR-tree [KS97], TV-tree =-=[LJF94]-=- or M-tree [CPZ97]. In general terms, these methods partition the data space, cluster the data according to the partitioning and use a hierarchical structure to prune the search space during querying.... |

157 | Beyond uniformity and independence: Analysis of r-trees using the concept of fractal dimension
- FALOUTSOS, KAMEL
- 1994
(Show Context)
Citation Context ... uniformity and independence assumptions are questionable in practice. However, it has been shown that the concept of fractal dimensionality is in fact a useful measure for the performance of R-trees =-=[FK94]-=-. In some sense, the fractal dimensionality represents the inherent dimensionality of the problem and can be somewhat smaller than the actual dimensionality. Declustering techniques proposed by [KF92]... |

149 | Data structures for range searching - Bentley, Friedman - 1979 |

71 | Content-based image indexing
- Chiueh
- 1994
(Show Context)
Citation Context ...rectly to cluster objects together (R-tree [Gut84], R*-tree [BKSS90], Xtree [BKK96], SR-tree [KS97], TV-tree [LJF94]) whereas other methods only use distances between objects (M-tree [CPZ97], vp-tree =-=[Chi94]-=-). The shape of a page can be either a hyper cube (R-tree, R*-tree, X-tree), a hyper sphere (TV-tree, M-tree, vp-tree) or may vary between these two shapes (SR-tree). However, while these methods gene... |

69 | C.: Parallel R-trees
- Kamel, Faloutsos
(Show Context)
Citation Context ...een objects are given and these distances are expensive to compute, M-tree may still save a lot of distance computations during nearest neighbor search. 3 2.2 Parallelization of Tree Based Methods In =-=[KF92]-=-, a parallel version of R-tree is presented. The goal of this approach is to maximize the throughput, which results in the following two requirements for a parallel index method: ffl Store nodes that ... |

32 | Spectral covariance and fuzzy regions for image indexing
- Stricker, Dimai
- 1997
(Show Context)
Citation Context ...ts. Most often, these features are considered as points in high-dimensional vector spaces. Typically, the dimensionality of such features is rather small, between 4 and 8 in [FSN + 95, SO95] or 45 in =-=[SD96]-=-, but can become quite large like 315 in [Dim97] and over 900 in [Csi95, Csi97]. The similarity of two objects is then determined as the distance between two feature vectors. As such, similarity searc... |

14 |
An approximation based data structure for similarity search
- Weber, Blott
- 1997
(Show Context)
Citation Context ... low-dimensional vector spaces, their search performance deteriorates as soon as dimensionality exceeds 10. This problem has become known as the dimensional curse problem and has been investigated in =-=[WB97]-=- from a theoretical point of view. A new approach, the so called VA-File [BW97, WB97], overcomes this problem by not partitioning the data space, but rather following the signature approach of assigni... |

10 | A Vector-Approximation File for Similarity Search in High-Dimensional Vector Spaces - Blon, Weber - 1997 |

5 |
Differences of global features for region indexing
- Dimai
- 1997
(Show Context)
Citation Context ...s points in high-dimensional vector spaces. Typically, the dimensionality of such features is rather small, between 4 and 8 in [FSN + 95, SO95] or 45 in [SD96], but can become quite large like 315 in =-=[Dim97]-=- and over 900 in [Csi95, Csi97]. The similarity of two objects is then determined as the distance between two feature vectors. As such, similarity search corresponds to a nearest-neighbor search withi... |

4 | Information extraction by local density analysis: A contribution to content-based management of scientific data - Csillaghy - 1997 |

3 | Retrieving information from digital solar radio spectrograms - Csillaghy - 1995 |

1 | Partitioned signature manager. Student project (in german), Institute of Information Systems - Bachmann - 1996 |

1 | Similarity search in a hospital information system---implementation of a scalable, parallelizable signature manager for similarity search. Diploma thesis (in german), Institute of Information Systems - Paschoud - 1997 |

1 | Signature manager for similarity search. Diploma thesis (in german - Schraner - 1996 |