I 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 techn

I 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.