Reinforcement learning with unsupervised auxiliary tasks

[joaorico]

https://deepmind.com/blog/reinforcement-learning-unsupervised-auxiliary-tasks/

[tener]

> On Atari the agent now achieves on average 9x human performance

Very impressive. I guess the human limit has to do with humans being limited about number of things to track at once? I wonder if they can apply this to optimizing the traffic lights in a big city.

[TrevorReznk]

Can't wait to see how it will handle Starcraft 2.