P. Zezula, P. Savino, G. Amato, and F. Rabitti. Approximate similarity retrieval with M-trees. The VLDB Journal, 4:275--293, 1998.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....also based on the retrieved set reduction idea, where hashing is used instead of a multi dimensional index tree to identify the retrieved data set. The disadvantage of this approach, however, is that # needs to be known in advance, and some preprocessing, which is exponential in 1 #, is needed. In [49], approximate similarity retrieval with M trees has been proposed. In [21, 23, 32] retrieved set reduction is achieved by clustering the data, and for each query, retrieving only a few clusters which are stored sequentially in the disk. Dimensionality Reduction: In [16] a dimensionality ....

....a fairly good answer set, although points g and h are likely to be uninteresting because of their distance to the query. q . a b . h . g . Figure 9: Inaccuracy of ranking based measures. An alternative quality measure was introduced in [19] and [49]. We now introduce a slightly more generalized version of that measure, to be used in our performance comparisons of various approximate k NN searching methods. Suppose the approximate k NN searching algorithm returns the result set a1 , a2, a k # and the actual (or golden) result set ....

P. Zezula, P. Savino, G. Amato, and F. Rabitti. Approximate similarity retrieval with M-trees. The VLDB Journal, 4:275--293, 1998.

....of points within MBRs. In the context of approximate searching mechanism a natural question arises: Is it possible to achieve faster response time if we allow some false dismissals Recently, various approaches in different domains have developed effective algorithms for approximate searching [4, 3, 22, 14, 41, 44, 17]. Most of these techniques specifically focus on ffl nearest neighbor queries. The ffl NN is defined as finding a neighbor of the query point within a factor of (1 ffl) of the distance to the true nearest neighbor. In [14] an algorithm was proposed in which the error bound ffl can be exceeded ....

P. Zezula, P. Savino, G. Amato, and F. Rabitti. Approximate similarity retrieval with M-trees. The VLDB Journal, 4:275--293, 1998. 20

....of points within MBRs. In the context of approximate searching mechanism a natural question arises: Is it possible to achieve faster response time if we allow some false dismissals Recently, various approaches in different domains have developed effective algorithms for approximate searching [4, 3, 21, 14, 39, 42, 16]. Most of these techniques specifically focus on ffl nearest neighbor queries. The ffl NN is defined as finding a neighbor of the query point within a factor of (1 ffl) of the distance to the true nearest neighbor. In [14] an algorithm was proposed in which the error bound ffl can be exceeded ....

P. Zezula, P. Savino, G. Amato, and F. Rabitti. Approximate similarity retrieval with M-trees. The VLDB Journal, 4:275--293, 1998. 17

....measure, however, is not able to capture inversions in the result (e.g. when image I s is ranked higher than image I s 0 in the approximate result and lower in the exact result) since no di erence between ranking of objects in the approximate and in the exact result is taken into account. In [ZSAR98] the precision of approximation measure P is introduced, which is de ned as: P = P k i=1 i rank(i) k (10) P , therefore, measures the relative error in ranking for all the objects in the approximate result. This measure, however, relies on the assumption that i rank(i) thus no ....

P. Zezula, P. Savino, G. Amato, and F. Rabitti. Approximate similarity retrieval with M-trees. The VLDB Journal, 7(4):275-293, 1998.

....hashing scheme proposed by Indyk and Motwani [25] The key idea is to use hash functions such that the probability of collision is much higher for objects that are close to each other than for those that are far apart. Approximate search has also been applied to tree like structures. Recent work [14, 46] shows that if one can tolerate 0 relative error with a ffi confidence factor, one can improve the performance of M tree by 1 2 orders of magnitude. Although an approximate nearest neighbor search can reduce the search space significantly, its recall can be low. This is because the ....

P. Zezula, P. Savino, G. Amato, and F. Rabitti. Approximate similarity retrieval with m-trees. VLDB Journal, (4):275--293, December 1998.

....[6] the latter only requiring that the distance is a metric, thus suitable even when no adequate vector representation for the features is possible. To obviate this unpleasant situation, several approximate solutions have been proposed that allow errors in the result in order to reduce costs [1, 16]. In this paper we propose a probabilistic approach, in which a NN query can specify two additional parameters: the accuracy ffl allows for a certain relative error, and the confidence ffi guarantees, with probability at least (1 Gamma ffi) that ffl will not be exceeded. This generalizes both ....

....The quality of the result of such algorithms is typically evaluated by the effective (relative) error, ffl eff , defined as: ffl eff = r r q Gamma 1 (1) where r r q is the distance between q and the approximate NN returned by the algorithm. Approximately correct NN (AC NN) algorithms [1, 16] use an accuracy parameter (relative error) ffl, to bound ffl eff , i.e. they return a point p 0 (called a (1 ffl) approximate NN) for which: d(q; p 0 ) 1 ffl)r q (2) surely holds. The optimal algorithm for C NN queries can be easily adapted to support AC NN queries, by pruning all ....

[Article contains additional citation context not shown here]

P. Zezula, P. Savino, G. Amato, and F. Rabitti. Approximate similarity retrieval with M-trees. The VLDB Journal, 7(4):275--293, 1998.

No context found.

P. Zezula, P. Savino, G. Amato, and F. Rabitti. Approximate similarity retrieval with M-trees. The VLDB Journal, 4:275--293, 1998.

No context found.

P. Zezula, P. Savino, G. Amato, and F. Rabitti, "Approximate similarity retrieval with m-trees," VLDB Journal, vol. 7, no. 4, pp. 275--293, December 1998.

No context found.

Pavel Zezula, Pasquale Savino, Giuseppe Amato, and Fausto Rabitti. Approximate similarity retrieval with M-trees. The VLDB Journal, 7(4):275-293, 1998.

No context found.

Pavel Zezula, Pasquale Savino, Giuseppe Amato, and Fausto Rabitti. Approximate similarity retrieval with M-trees. The VLDB Journal, 7(4):275-293, 1998.

No context found.

P. Zezula, P. Savino, G. Amato, and F. Rabitti. Approximate similarity retrieval with M-trees. VLDB Journal, 7(4):275--293, 1998.

No context found.

P. Zezula, P. Savino, G. Amato, and F. Rabitti. Approximate similarity retrieval with M-trees. VLDB Journal, 7(4):275--293, 1998.

No context found.

Pavel Zezula, Pasquale Savino, Giuseppe Amato, and Fausto Rabitti. Approximate similarity retrieval with M-trees. The VLDB Journal, 7(4):275-293, 1998.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC