2 ms·
April 1 jokes already? Storing floats as bit, great idea!
by rurban 3y ago
April 1 jokes already? Storing floats as bit, great idea!
- riku_iki 3y agobit embeddings are different to float, they are trained separately.
- marginalia_nu 3y agoI think the the technique is best understood as a special case of random projection. It's well understood that there are a class of dimensional reduction techniques that preserve distances well (via the J-L lemma), owing to the countrerintuitive fact that most high dimensional vectors are nearly orthogonal.