Learning Fast with No Goals - VISR Explained

VISR is a paper about rapid generalization to new tasks in Reinforcement Learning (RL). The full paper was released by DeepMind in 2020 and is called Fast Task Inference with Variational Intrinsic Successor Features. It uses successor features and goal-conditioned policies to rapidly adapt to new tasks after learning within the no-reward regime of RL.

Link to paper: https://openreview.net/forum?id=BJeAHkrYDS

explained