an open-source infrastructure paradigm for building parallel reinforcement learning algorithms in TensorFlow, allowing new algorithms to be developed and trained efficiently