[27] Danqi Chen - Neural Reading Comprehension and Beyond
The Thesis Review
00:00
The Challenge of Defining the Right Problem
Different data sets actually have different patterns. And the models actually, the big neural networks are actually very good at like capturing all these patterns. But still we know that models cannot really generalize very well from one data set to another. We need to put more effort into understanding and analyzing what these models actually try to learn. Then we actually have a model that can actually deploy in the real world. Dr. QA: Could you just talk through like the backstory of working on this? Yeah, kind of like what problem was it solving?
Transcript
Play full episode
