A fast nearest neighbor method using empirical marginal distribution

Abstract

Unfortunately there is no essentially faster algorithm than the brute-force algorithm for the nearest neighbor searching in high-dimensional space. The most promising way is to find an approximate nearest neighbor in high probability. This paper describes a novel algorithm that is practically faster than most of previous algorithms. Indeed, it runs in a sublinear order of the data size. © 2009 Springer Berlin Heidelberg.