The Why and How of Nonnegative Matrix Factorization — HN submission

Wed, 06 Mar 2019 07:05:00 GMT

Why and How of Non-Negative Matrix Factorization

2 points; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

The Why and How of Nonnegative Matrix Factorization — mentioned by FabHK

Mon, 18 Feb 2019 15:45:05 GMT

Interesting. The problem of finding a low-rank-approximation (in Frobenius norm) to a given matrix is well understood and solved in reasonable time (truncated SVD, as stated in this article). However, the particular characteristic of this NMF approach seems to be that we require W, H >= 0 (and that's what makes it harder), but it's not clear why, what purpose it serves. The original paper linked [1] mentions that it makes interpretation easier (the vectors of W can be interpreted as facial features, or a form of "eigenface"), but it's not clear to me why you can't take a vector with negative…

Paper and discussion history. A mention is not an endorsement.

The Why and How of Nonnegative Matrix Factorization · HackerPapers discussions

The Why and How of Nonnegative Matrix Factorization — HN submission

The Why and How of Nonnegative Matrix Factorization — mentioned by FabHK