All-Pairs Nearest Neighbor Search on Manycore Systems
Lawrence Cayton
NIPS workshop on learning on cores, clusters, and clouds 2010.

Abstract

In this note, we describe a data structure for similarity search in metric spaces that is simple to parallelize effectively. We present results of a GPU implementation of our data structure for the all-pairs nearest neighbor problem, which frequently arises in semi- and un-supervised learning algorithms.

[Edit]