Deep-Q learning using Atari games as the environment :). I don’t do anything at all with that now as a computational linguist but I loved it as a way to talk about what it means to learn a policy gradient (or way more simply, how to react to things) in a continuous changing world :)