Skip to content

Comment on Classifying all of the pdfs on the internetparent

Comments

Don't most statistical techniques rely on specific structure in the spaces containing the objects they operate on, in order to be useful?

Embeddings have structure, or they wouldn't be very useful. E.g. cosine similarity works because (many) embeddings are designed to support it.

Oh, that should have been obvious. Thank you for explaining.

AboutSource Built by g1lg1l

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