Max Schwarzer
TalkRL: The Reinforcement Learning Podcast
00:00
How to Train a Big Network Effectively With No Prior Knowledge
BBF is inspired by work from, I would say, the self-supervised learning literature. It's interesting to me that with BBF, you're getting to human level learning times roughly ish and yet exceeding that in many cases. But of course, humans come in into a game with a lot of prior knowledge about things like what objects are and what actions mean. And so our agents don't have any of this background knowledge.
Transcript
Play full episode
