https://hgpu.org/?p=5394
A CUDA-based parallel implementation of K-nearest neighbor algorithm