123 - Robust NLP, with Robin Jia
NLP Highlights
00:00
The Differences Between High and Low Frequency Observations in the Language Setting
There's been this interesting observation that there's kind of a distinction between high frequency and low frequency perturbations. And so if you try to make your model robust to high frequency perturbation, it'll often help for other highfrequency perturbations but then it'll be more sensitive to certain low frequency ones. So being robust to something doesn't mean you're not the robust to other things. Could you talk a little bit about training more robust models?
Transcript
Play full episode
