Re:Separability, SVD and low-rank approximation of 2D image ...
by noelwelsh on Monday March 02, 2020 @04:30AM
(#22462941)
Good stuff. A lot of clustering methods can be viewed as matrix factorization (see, e.g., https://arxiv.org/abs/1512.07548) so it would be interesting to see if you could spend more computation on the creating the low dimensional approximation to achieve a better result. E.g. perhaps optimizing the absolute error (L1 norm) would be better than the squared error (L2 norm)?
[ Parent
| on HN ]
There are no comments on this comment yet.