[24] Martin Arjovsky - Out of Distribution Generalization in Machine Learning
The Thesis Review
00:00
Is There a Difference Between Overparameterized and Not?
If you don't have an overparameterized problem IRM will do the job pretty well. You see this in most applications that IRM works on are kind of a bad flavor. This is something that we cannot address right now but it and it's really hard in the sense that it mixes what you discussed of like the implicit bias of sgd.
Transcript
Play full episode
