123 - Robust NLP, with Robin Jia
NLP Highlights
00:00
The Importance of Reporting Robustness
The active learning still creates a relatively balanced data set I think it's more that you're shifting. So when you have this extreme label imbalance there's like tons of negative examples most of them should at least be very easy for a model right. If you just kind of randomly sample them you're not going to see enough of the interesting hard cases. In fact the learning lets you say like I'm going to select the prioritize the cases that are hard and yeah and still collect but still collect kind of relatively balanced dataSet.
Transcript
Play full episode
