https://hgpu.org/?p=4327
Parallel implementation of artificial neural network training