[08] He He - Sequential Decisions and Predictions in NLP
The Thesis Review
00:00
How Machine Learning Interacts With Structure Prediction
Imitation learning is basically studies how do you learn from expert demonstrations. Once you go off the article pass, you make a mistake and now you no longer have article demonstrations. So what do you do in those cases? That's the key challenge or the core problem most imitation or narrow reasons trying to solve. And I think it's also nice because it allows you to reduce it to supervised learning. For example, in Dagger, you assume that you already have this classifier and then you're just changing the supervision signal. You could use reduction.
Transcript
Play full episode
