Spectral Approaches to Nearest Neighbor Search

被引:8
作者
Abdullah, Amirali [1 ]
Andoni, Alexandr
Kannan, Ravindran
Krauthgamer, Robert [2 ]
机构
[1] Univ Utah, Salt Lake City, UT 84112 USA
[2] Weizmann Inst Sci, Rehovot, Israel
来源
2014 55TH ANNUAL IEEE SYMPOSIUM ON FOUNDATIONS OF COMPUTER SCIENCE (FOCS 2014) | 2014年
关键词
ALGORITHM;
D O I
10.1109/FOCS.2014.68
中图分类号
TP301 [理论、方法];
学科分类号
081202 ;
摘要
We study spectral algorithms for the high-dimensional Nearest Neighbor Search problem (NNS). In particular, we consider a semi-random setting where a dataset is chosen arbitrarily from an unknown subspace of low dimension, and then perturbed by full-dimensional Gaussian noise. We design spectral NNS algorithms whose query time depends polynomially on the dimension and logarithmically on the size of the point set. These spectral algorithms use a repeated computation of the top PCA vector/subspace, and are effective even when the random-noise magnitude is much larger than the interpoint distances. Our motivation is that in practice, a number of spectral NNS algorithms outperform the random-projection methods that seem otherwise theoretically optimal on worst-case datasets. In this paper we aim to provide theoretical justification for this disparity. The full version of this extended abstract is available on arXiv.
引用
收藏
页码:581 / 590
页数:10
相关论文
共 50 条
[1]   Problems and results in extremal combinatorics - I [J].
Alon, N .
DISCRETE MATHEMATICS, 2003, 273 (1-3) :31-53
[2]  
Andoni A, 2006, ANN IEEE SYMP FOUND, P459
[3]  
Andoni A, 2006, ANN IEEE SYMP FOUND, P449
[4]  
[Anonymous], P S THEOR COMP STOC
[5]  
[Anonymous], P ACM S COMP GEOM SO
[6]  
[Anonymous], ACM T ALGORITHMS
[7]  
[Anonymous], SODA
[8]  
[Anonymous], COMP VIS ICCV 2011 I
[9]  
[Anonymous], P ACM SIAM S DISCR A
[10]  
[Anonymous], 43 ANN IEEE S FDN CO