123 - Robust NLP, with Robin Jia
NLP Highlights
00:00
The Problem With Local Rewrite Operations in a Movie Review
The difference that you found between the really high standard 90% accuracy and 8% robustness accuracy kind of naturally brings about the question, can we fix this? Like are there ways of making models more robust to these kinds of perturbations? And I know you've worked on this. Could you talk a little bit about how your work went about this problem and what you found?
Transcript
Play full episode
