Max Schwarzer
TalkRL: The Reinforcement Learning Podcast
00:00
The Importance of Model-Based Learning
The SPR self-supervised objective in BBF helps you to learn useful representations of the world. But we never use it for planning. It's simply actually doing this objective. And I think we would expect to see this in general in a lot of cases, right? If I'm reconstructing,. even just the simplest thing you could think of, reconstructing the next frame, I absolutely expect that to get you better representations than purely doing reinforcement learning. That interplay is incredibly powerful.
Transcript
Play full episode
