Accelerating Reinforcement Learning through GPU Atari Emulation