Comment on A polynomial autoencoder beats PCA on transformer embeddingsComments−yobbo4moMy 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.−ekjhgkejhgk4moI think that kernel PCA is a strict superset of PCA. That would make it trivially true that it beats PCA.
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.