Episode 22: Archit Sharma, Stanford, on unsupervised and autonomous reinforcement learning
Generally Intelligent
00:00
Is Distribution Matching a Natural Learning Approach?
This is again a very natural idea. Like, as you were talking about earlier, as well as like humans learn to do something once and then just kind of keep doing it even if that's not the optimal thing to do. And so what you did was you had this Q-weighted adversarial learning. And you're using distribution matching, where you like, like match to agents prior experience. Can you explain how it works? Absolutely.
Transcript
Play full episode
