Tweeted By @DeepMind
How can we make #reinforcementlearning agents learn faster?
— DeepMind (@DeepMind) August 18, 2020
A new article published in @PNASNews proposes a divide-and-conquer approach to RL that allows an agent to combine the solution of known tasks to quickly solve new ones: https://t.co/mcEPNBDLzJ pic.twitter.com/Vek5HrvOXe