NLP Highlights cover image

117 - Interpreting NLP Model Predictions, with Sameer Singh

NLP Highlights

00:00

How to Compute an Influence Function

We didn't talk very much about how exactly the influence function works. Can you give a high level description of what's going on when you're computing influence functions? So if I were to change my parameter number one, seven, two, three, how much will it affect the output? Once you have this information, you can go back and see, okay, for every training data in my every training point in my training data,. If I was to remove that training instance, how much would theparametriest change? And that's in some sense another sort of gradient style update. This is an approximation of what we do if you were to train to convergence.

Transcript
Play full episode

The AI-powered Podcast Player

Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!
App store bannerPlay store banner
Get the app