Parallel Construction of Approximate kNN Graph

Abstract

Building k-nearest neighbor (kNN) graphs is a necessary step in such areas as data mining and machine learning. So in this paper, we attempt to study the kNN furthermore, we first propose a parallel algorithm for approximate kNN graph construction and then apply the kNN graph to the application of clustering. Experiments show that our MPI/OpenMP mixed mode codes can make the construction of approximate kNN graph faster and make the parallelization and implementation easier. Finally, we compare the results of agglomerative clustering methods by using our parallel algorithm to illustrate the applicability of this method.