Chelsea Finn on Meta Learning & Model Based Reinforcement Learning
The Gradient: Perspectives on AI
00:00
The Impact of Meta-Learning on Networking
The idea is to try to learn an initial set of weights, such that if you fine-tune for a small number of steps, you can solve the task very, very quickly. And at the beginning, it really wasn't obvious that it was going to work. It really wasn't clear that you could actually get very far with the single gradient step or a few gradient steps. But it turns out that you can, and that's what those initial experiments were really trying to test.
Transcript
Play full episode
