Thomas Lahore Dec 8
This is a SUPER interesting idea! Training Agents using Upside-Down Reinforcement Learning "in upside-down RL ... the roles of actions and returns are switched."