Been Kim: Interpretable Machine Learning
The Gradient: Perspectives on AI
00:00
How to Improve the Design of a Neural Network?
The difficulty in both the generative model work and the causal work is the following. How do you know that it's faithfully drawing in and not just pretending to draw, right? So that's difficulty there. The difficulty in causal T-cap is that you need a causal graph that you trust and you believe that there are no other unknown comp buildings. That just is simply an assumption. You have to have in causal inference that simply you never can validate. If you have wrong causal graph, then any conclusion that you see could be incorrect and you wouldn't know it. And understanding limitation is one of the really important ways to move forward.
Transcript
Play full episode
