Adaptive cluster distance bounding

For Diploma, BE, ME, M Tech, BCA, MCA, PHD Project Guidance, Please Visit: www.ocularsystems.in Or
Call Us On 7385665306

Adaptive Cluster Distance Bounding for High
Dimensional Indexing+
Abstract:

We consider approaches for similarity search in correlated, high-dimensional data
sets, which are derived within a clustering framework. We note that indexing by
“vector approximation” (VA-File), which was proposed as a technique to combat the
“Curse of Dimensionality,” employs scalar quantization, and hence necessarily
ignores dependencies across dimensions, which represents a source of sub
optimality. Clustering, on the other hand, exploits inter dimensional correlations
and is thus a more compact representation of the data set. However, existing
methods to prune irrelevant clusters are based on bounding hyperspheres and/or
bounding rectangles, whose lack of tightness compromises their efficiency in exact
nearest neighbor search. We propose a new cluster-adaptive distance bound based
on separating hyperplane boundaries of Voronoi clusters to complement our cluster
based index. This bound enables efficient spatial filtering, with a relatively small
preprocessing storage overhead and is applicable to euclidean and Mahalanobis
similarity measures. Experiments in exact nearest-neighbor set retrieval, conducted
on real data sets, show that our indexing method is scalable with data set size and
data dimensionality and outperforms several recently proposed indexes. Relative to
the VA-File, over a wide range of quantization resolutions, it is able to reduce
random IO accesses, given (roughly) the same amount of sequential IO operations,
by factors reaching 100X and more.

Ocular Systems, Shop No:1, Swagat Corner Building, Near Narayani Dham Temple, Katraj, Pune-46

E-Mail: info@ocularsystems.in

Call Us On 7385665306

Existing System:

However, existing methods to prune irrelevant clusters are based on bounding
hyperspheres and/or bounding rectangles, whose lack of tightness compromises
their efficiency in exact nearest neighbor search.

Spatial queries, specifically nearest neighbor queries, in high-dimensional spaces
have been studied extensively. While several analyses have concluded that the
nearest neighbor search, with Euclidean distance metric, is impractical at high
dimensions due to the notorious “curse of dimensionality”, others have suggested
that this may be over pessimistic. Specifically, the authors of have shown that
what Determines the search performance (at least for R-tree-like structures) is the
intrinsic dimensionality of the data set and not the dimensionality of the address
space (or the embedding dimensionality).

We extend our distance bounding technique to the Mahalanobis distance metric,
and note large gains over existing indexes.

Proposed System:

We propose a new cluster-adaptive distance bound based on separating hyperplane
boundaries of Voronoi clusters to complement our cluster based index. This bound
enables efficient spatial filtering, with a relatively small pre-processing storage
overhead and is applicable to Euclidean and Mahalanobis similarity measures.
Experiments in exact nearest-neighbor set retrieval, conducted on real data-sets,
show that our indexing method is scalable with data-set size and data
dimensionality and outperforms several recently proposed indexes.



Call Us On 7385665306

we outline our approach to indexing real high-dimensional data-sets. We focus on
the clustering paradigm for search and retrieval. The data-set is clustered, so that
clusters can be retrieved in decreasing order of their probability of containing
entries relevant to the query.

We note that the Vector Approximation (VA)-file technique implicitly assumes
independence across dimensions, and that each component is uniformly distributed.
This is an unrealistic assumption for real data-sets that typically exhibit significant
correlations across dimensions and non-uniform distributions. To approach
optimality, an indexing technique must take these properties into account. We
resort to a Voronoi clustering framework as it can naturally exploit correlations
across dimensions (in fact, such clustering algorithms are the method of choice in
the design of vector quantizers). Moreover, we show how our clustering procedure
can be combined with any other generic clustering method of choice (such as
BIRCH ) requiring only one additional scan of the data-set. Lastly, we note that the
sequential scan is in fact a special case of clustering based index i.e. with only one
cluster.
Several index structures exist that facilitate search and retrieval of multi-
dimensional data. In low dimensional spaces, recursive partitioning of the space
with hyper-rectangles hyper-spheres or a combination of hyper-spheres and
hyper-rectangles have been found to be effective for nearest neighbor search and
retrieval. While the preceding methods specialize to Euclidean distance (l2 norm),
M-trees have been found to be effective for metric spaces with arbitrary distance
functions (which are metrics).

Such multi-dimensional indexes work well in low dimensional spaces, where they
outperform sequential scan. But it has been observed that the performance
degrades with increase in feature dimensions and, after a certain dimension
threshold, becomes inferior to sequential scan. In a celebrated result,



Call Us On 7385665306

Weber et. Al have shown that whenever the dimensionality is above 10, these
methods are outperformed by simple sequential scan. Such performance
degradation is attributed to Bellman’s ‘curse of dimensionality’, which refers to the
exponential growth of hyper-volume with dimensionality of the space.

Module Description:

1. A New Cluster Distance Bound
2. Adaptability to Weighted Euclidean or Mahalanobis Distances
3. An Efficient Search Index
4. Vector Approximation Files
5. Approximate Similarity Search

A New Cluster Distance Bound

Crucial to the effectiveness of the clustering-based search strategy is efficient
bounding of query-cluster distances. This is the mechanism that allows the
elimination of irrelevant clusters. Traditionally, this has been performed with
bounding spheres and rectangles. However, hyperspheres and hyperrectangles are
generally not optimal bounding surfaces for clusters in high dimensional spaces. In
fact, this is a phenomenon observed in the SR-tree, where the authors have used a
combination spheres and rectangles, to outperform indexes using only bounding
spheres (like the SS-tree) or bounding rectangles (R∗-tree).

The premise herein is that, at high dimensions, considerable improvement in
efficiency can be achieved by relaxing restrictions on the regularity of bounding
surfaces (i.e., spheres or rectangles). Specifically, by creating Voronoi clusters,
withpiecewise-linear boundaries, we allow for more general convex polygon



Call Us On 7385665306

structures that are able to efficiently bound the cluster surface. With the
construction of Voronoi clusters under the Euclidean distance measure, this is
possible. By projection onto these hyperplane boundaries and complementing with
the cluster-hyperplane distance, we develop an appropriate lower bound on the
distance of a query to a cluster.

Adaptability to Weighted Euclidean or Mahalanobis Distances

While the Euclidean distance metric is popular within the multimedia indexing
community it is by no means the “correct” distance measure, in that it may be a
poor approximation of user perceived similarities. The Mahalanobis distance
measure has more degrees of freedom than the Euclidean distance and by proper
updation (or relevance feedback), has been found to be a much better estimator of
user perceptions and more recently) . We extend our distance bounding technique
to the Mahalanobis distance metric, and note large gains over existing indexes.

An Efficient Search Index

The data set is partitioned into multiple Voronoi clusters and for any kNN query, the
clusters are ranked in order of the hyperplane bounds and in this way, the
irrelevant clusters are filtered out. We note that the sequential scan is a special
case of our indexing, if there were only one cluster. An important feature of our
search index is that we do not store the hyperplane boundaries (which form the
faces of the bounding polygons), but rather generate them dynamically, from the
cluster centroids. The only storage apart from the centroids are the cluster-
hyperplane boundary distances (or the smallest cluster-hyperplane distance). Since
our bound is relatively tight, our search algorithm is effective in spatial filtering of



Call Us On 7385665306

irrelevant clusters, resulting in significant performance gains. We expand on the
results and techniques initially presented in , with comparison against several
recently proposed indexing techniques.

Vector Approximation Files

A popular and effective technique to overcome the curse of dimensionality is the
vector approximation file (VA-File). VA-File partitions the space into hyper-
rectangular cells, to obtain a quantized approximation for the data that reside inside
the cells. Non-empty cell locations are encoded into bit strings and stored in a
separate approximation file, on the hard-disk. During a nearest neighbor search,
the vector approximation file is sequentially scanned and upper and lower bounds
on the distance from the query vector to each cell are estimated. The bounds are
used to prune irrelevant cells. The final set of candidate vectors are then read from
the hard disk and the exact nearest neighbors are determined. At this point, we
note that the terminology “Vector Approximation” is somewhat confusing, since
what is actually being performed is scalar quantization, where each component of
the feature vectors separately and uniformly quantized (in contradistinction with
vector quantization in the signal compression literature).

VA-File was followed by several more recent techniques to overcome the curse of
dimensionality. In the VA+-File, the data-set is rotated into a set of uncorrelated
dimensions, with more approximation bits being provided for dimensions with
higher variance. The approximation cells are adaptively spaced according to the
data distribution. Methods such as LDR and the recently proposed non-linear
approximations aim to outperform sequential scan by a combination of clustering
and dimensionality reduction. There also exist a few hybrid methods, such as the A-
Tree, and IQ-Tree, which combine VA-style approximations within a tree based
index.
Approximate Similarity Search



Call Us On 7385665306

Lastly, it has been argued that the feature vectors and distance functions are often
only approximations of user perception of similarity. Hence, even the results of an
exact similarity search is inevitably perceptually approximate, with additional
rounds of query refinement necessary. Conversely, by performing an approximate
search, for a small penalty in

accuracy, considerable savings in query processing time would be possible.
Examples of such search strategies are MMDR probabilistic searches and locality
sensitive hashing .The reader is directed to for a more detailed survey of
approximate similarity search. The limits of approximate indexing i.e. the optimal
tradeoffs between search quality and search time has also been studied within an
information theoretic framework.

System Architecture:



Call Us On 7385665306

Hardware System Requirement

Processor - Pentium –III

Speed - 1.1 Ghz

RAM - 256 MB(min)

Hard Disk - 20 GB

Floppy Drive - 1.44 MB

Key Board - Standard Windows Keyboard

Mouse - Two or Three Button Mouse

Monitor - SVGA

S/W System Requirement

 Operating System : Windows 95/98/2000/NT4.0.

 Application Server : Tomcat6.0

 Front End : HTML, Java.

 Scripts : JavaScript.

 Server side Script : Java Server Pages.

 Database : Mysql.

 Database Connectivity : JDBC.



Adaptive cluster distance bounding

More Related Content

What's hot (19)

Similar to Adaptive cluster distance bounding (20)

More from Ocular Systems (6)

Recently uploaded (20)

Adaptive cluster distance bounding