Research on Parallelization of GPU-based K-Nearest Neighbor Algorithm

Hao Jiang,Yulin Wu
DOI: https://doi.org/10.1088/1742-6596/910/1/012006
2017-01-01
Journal of Physics Conference Series
Abstract:Based on the analysis of the K-Nearest Neighbor Algorithm, the feasibility of parallelization is studied from the steps of the algorithm, the operation efficiency and the data structure of each step, and the part of parallel execution is determined. A K-Nearest Neighbor Algorithm parallelization scheme is designed and the parallel G-KNN algorithm is implemented in the CUDA environment. The experimental results show that the K-Nearest Neighbor Algorithm has a significant improvement in efficiency after parallelization, especially on large-scale data.
What problem does this paper attempt to address?