114 - Behavioral Testing of NLP Models, with Marco Tulio Ribeiro
NLP Highlights
00:00
What Kinds of Tests Should I Write?
Ye: How should we think about what kinds of things to test? In your paper, you talk about sentiment analysis and reading comprehension. I would say one of these is much broader in scope than the other. Ye: There's no point in testing if my model understands sarcasm if he cannot handle this is a good movie yet. So i think we should start with what's likely to be possible right now. And if we can't do level one, three, that's great. That's where the matrix abstraction really helps.
Transcript
Play full episode
