The Deep-Mind results are excellent! Thank you! They used reinforcement learning to find their algorithms. In an amusing twist, my little article has a recurrent approach to Policy-Gradient that could sidestep some matrix multiplication in RL itself.
