[08] He He - Sequential Decisions and Predictions in NLP
The Thesis Review
The Importance of a Test-Time Distribution
For robustness, what's one outcome or goal that you would reach for? Would you want to model with some guarantee, which says it's guaranteed to be robust in X and YZ sense? Yeah, so I think if you want to guarantee, we have to think about what type of assumptions we want to put on the test-time distribution. If we make sure that the model work on this test distribution that differ from the training distribution in this way, we can say that the model is learning something that's not spurious. And the other scenario that's connected to this sequential decision-making or interactive learning setting that I think is interesting is maybe we could add a humor in this learning
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
