Isn't SVD a dimensionality reduction technique?

In week 2, as far as I understood, SVD was introduced as a dimensionality reduction technique. But then again, in week 4’s quiz, there is this question which asks about unsupervised drift detection techniques and the fact that they suffer from curse of dimensionality. We are supposed to select a list of dimensionality reduction techniques but selecting SVD results in an incorrect answer. Isn’t SVD a dimensionality reduction technique? What am I missing?

Since I didn’t get any replies, here’s a screenshot of the question and the reasoning why SVD is not an acceptable option which does not make sense to me:

{moderator edit - quiz answers removed}

Yes, it’s a slightly subtle point, but SVD is a tool that you can use to do dimensionality reduction. You can use it for other things as well. Or to put it a slightly different way: what SVD tells you is which dimensions can reasonably be removed, but it doesn’t actually remove them. That’s a separate step.

You could be forgiven for wondering if perhaps the person who wrote this question might have considered going to law school as an alternative career path. :laughing:

1 Like

Thanks. In that case, I would say including SVD here (in this question) is just simply misleading. Granted that now I know one more thing about SVD but unless this explanation (provided by you) is mentioned in the quiz, it’s just a bad question design.

Well, to be fair, they did give better answers to choose. PCA is really the dimensionality reduction technique and it uses SVD to figure out which eigenvectors are the ones that matter.