Chelsea Finn on Meta Learning & Model Based Reinforcement Learning
The Gradient: Perspectives on AI
00:00
The Future of Predictive Models
It's a pretty powerful idea, even if it's kind of difficult to think about physics and predict video. And the cool bit is, as you said, that all of the needs is this pre-learned model, and then it doesn't learn to do these new tasks. You just give it some specification of what you wanted to accomplish, and it just sort of plans, it looks ahead, what sequence of actions get it to there, and does it. So it doesn't need to learn really anything for a new task.
Transcript
Play full episode
