intelligent database systems lab n.y.u.s.t. i. m. fast exact k nearest neighbors search using an...

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.

Fast exact k nearest neighbors search using an orthogonal search tree

Presenter : Chun-Ping Wu　Authors :Yi-Ching Liaw, Maw-Lin Leou, Chien-Min Wu

PR 2010

國立雲林科技大學National Yunlin University of Science and Technology

1


N.Y.U.S.T.

I. M.Outline

Motivation Objective Methodology Experiments Conclusion Comments

2


N.Y.U.S.T.

I. M.Motivation

The finding process of k nearest neighbors for a query point using FSA(full search algorithm) is very time consuming.

Many algorithms want to reduce the computational complexity of the kNN finding process. Pre-created tree structure

3


N.Y.U.S.T.

I. M.Motivation

For a big PAT(Principal Axis Search), the computation time to evaluate boundary points and projection values will be large.

4


N.Y.U.S.T.

I. M.Objective

To reduce the computation time on evaluation boundary points and projection values in the kNN searching process for a query point.

The proposed method requires no boundary points and only little computation time on evaluating projection values in the kNN finding process.

5


N.Y.U.S.T.

I. M.Methodology

The OST(orthogonal search tree) algorithm OST construction process

K Nearest neighbors

search using the OST

6


N.Y.U.S.T.

I. M.Methodology

The OST construction process

7

1,2,3,4,5,6,7,8,9

1,2,3,4,5,6,7,8,9

1,2,3 4,5,6 7,8,9

1,2,3,4,5,6,7,8,9

1,2,3 4,5,6 7,8,9

1 2 3


N.Y.U.S.T.

I. M.Methodology

K nearest neighbors search

using the orthogonal search tree

8

1,2,3,4,5,6,7,8,9

1,2,3 4,5,6 7,8,9

1 2 3 4 5 6 7 8 9


N.Y.U.S.T.

I. M.Experiments

Example 1 Uniform Markov source

9


N.Y.U.S.T.

I. M.Experiments

10


N.Y.U.S.T.

I. M.Experiments

Example 2 auto-correlated data

11


N.Y.U.S.T.

I. M.Experiments

Example 3 Clustered Gaussian data

12


N.Y.U.S.T.

I. M.Experiments

Example 4 Data sets are codebook

generated using 6 real images.

13


N.Y.U.S.T.

I. M.Experiments

Example 5 Statlog data set.

14

34% 39%


N.Y.U.S.T.

I. M.Conclusion

1515

Experimental results show that the proposed method always spends less computation time to find the kNN for a query point than the other methods.

The proposed method will find the same results as those of the FSA(full search algorithm).


N.Y.U.S.T.

I. M.Comments

1616

Advantage To reduce the computation of the kNN finding process.

Drawback Lack of illustrations

Application Classification

intelligent database systems lab n.y.u.s.t. i. m. fast exact k nearest neighbors search using an...

Documents