Skip to content

Comment on A polynomial autoencoder beats PCA on transformer embeddings

Comments

My understanding after scanning the code examples is the technique expands the dimensionality of each data point with a set consisting of the quadratic coefficients of its existing dimensions. I thought it sounded like kernel PCA.

I think that kernel PCA is a strict superset of PCA. That would make it trivially true that it beats PCA.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.