Parallel k Nearest Neighbor Graph Construction Using Tree-Based Data Structures

Construction of a nearest neighbor graph is often a necessary step in many machine learning applications. However, constructing such a graph is computationally expensive, especially when the data is of high dimensionality. In this work, we focus on the use of two tree structures, k-d trees and ball trees, to implement nearest neighbor graph construction. We present parallel implementations of nearest neighbor graph construction using such tree structures, with parallelism provided by OpenMP and the Galois framework. Our results show that kd-trees are faster when the number of dimensions is small (N >> 2), ball trees on the other hand scale well will the number of dimensions. Our Galois implementation with maxspread was faster than OpenMP for data that satisfied the kd-tree requirement, irrespective of number of threads. OpenMP ball trees were faster than Galois kd-tree for datasets with large dimension, irrespective of number of threads.