This Week in AI, February 15th, 2018 – Udacity Inc – Medium

Alex Irpan, a software engineer at Google, wrote an excellent article on the current difficulties of getting deep reinforcement learning to work. For example, even after weeks of optimizing hyperparameters and explotation-exploration rates, these models are still highly sensitive to initial conditions. A 30% failure rate is seen as "working." Irpan makes the argument that most attempts with deep RL fail but no one talks about it publicly, we only see the few cases where the problems are simplified enough to be feasible. This is still a new field - the breakthrough Atari DQN paper was published only 3 years ago - so there is plenty of room for more research and advancement.