https://hgpu.org/?p=15472
Speeding Up Reinforcement Learning with Graphics Processing Units