For those of you who are only interested in DCCA section: 14:38
@brandomiranda6703
2 жыл бұрын
btw, I really appreciated the "functional" view of CCA. That was especially helpful to me.
@brandomiranda6703
2 жыл бұрын
I find the comment of the data views a little confusing. I have never seen that point to view. Perhaps that's why it's confusing to me. The way I understand that after further reading is that in reality general functional CCA finds two functions that correlate a collection of random variables x1, x2 (where the coordinates are random variables). The random variables might have a correlation or not. But I guess in Galen Andrews's work he is has applied it to data that comes form the same source - why I think he refers them as views. Essentially, that assumption makes it clear that the r.v.s should be correlated since the data is the same - i.e. the generation of data is from the same source but it generates different views i.e. different r.v.s. So the assumption (from the example) is that there *must* be some correlation, due to the problem setting. Which is fine, just trying to understand the general framework, application, etc. Good video in general! Thanks for sharing.
@sarahooshmand1830
2 жыл бұрын
Awesome ! Thanks for sharing
@InfinityDz
2 жыл бұрын
What is the gradient for the frobenius based error function relative to W and b? It's not in the paper and I couldn't find it in the references given
Пікірлер: 6