Max Schwarzer
TalkRL: The Reinforcement Learning Podcast
00:00
The Future of Search
The Asian is learning more of the mechanism. A lot of our algorithms are these certain tricks that we discover. But he was saying, what are the chances that we're going to really discover all these tricks manually? He put that at a very low probability and so we need to find ways to discover these things. Where for meta-RL to be really impactful, I think you have to come up with something that has the same property. It gets distilled down in some way that lets other people who don't have your agent still benefit from it later on.
Transcript
Play full episode
