Google's DeepMind AI improves learning speed and performance through UNREAL (UNsupervised REinforcement and Auxiliary Learning) - autonomously learning how actions affect what is seen, to predict the onset of immediate rewards from a short historical context - similar to the way animals dream.

See the source article by following the link below:
Google's DeepMind AI improves learning speed and performance through UNREAL (UNsupervised REinforcement and Auxiliary Learning) - autonomously learning how actions affect what is seen, to predict the onset of immediate rewards from a short historical context - similar to the way animals dream.