[24] Martin Arjovsky - Out of Distribution Generalization in Machine Learning
The Thesis Review
00:00
Is There a Causality in Graphs?
I didn't know anything about causality at the time I just knew that Leon was very excited on it which was enough for me to think that perhaps it is important. Six months after this internship we tried a lot of different things nothing worked but also nothing makes sense. The thing that remained the same when you intervene was the conditional distribution from the causes to to the target and then came the idea of finding features such that the conditional probability from those features to why remain the same across different data sets.
Transcript
Play full episode
