c*******a
发帖数: 1879 | 1 什么叫low rank matrix factorization?
应用场景？ |
B*******e
发帖数: 30 | 2 就是一个estimate matrix structural parameter subjected to rank = desired
rank 的问题 |
k****z
发帖数: 1863 | 3 瞎扯一下
就是把一个矩阵分解成几个最简单的矩阵乘积
想象成空间，就是一个复杂的空间变化分解成几个最简单空间变化的组合 |
c*******a
发帖数: 1879 | 4 你不行啊 没有回答实质问题啊
【在 B*******e 的大作中提到】
: 就是一个estimate matrix structural parameter subjected to rank = desired
: rank 的问题
|
h*******i
发帖数: 4386 | 5 svd?
【在 c*******a 的大作中提到】
: 什么叫low rank matrix factorization?
: 应用场景？
|
c*******a
发帖数: 1879 | 6 你数学功底太差
呼唤时崩姐
【在 h*******i 的大作中提到】
: svd?
|
C**********e
发帖数: 23303 | 7 大规模电力系统里的暂稳态分析潮流计算
【在 c*******a 的大作中提到】
: 什么叫low rank matrix factorization?
: 应用场景？
|
w*******e
发帖数: 17 | 8 Louzhu should study some basic ideas of Singular Value Decomposition.
Very often, low-rank matrix factorization means the process of finding a low
-rank matrix to approximate your original matrix.
If you use Frobenius norm to find the best approximation, i.e., minimize
over
|| D -\hat(D) ||_F subject to rank(\hat(D))
then you can get the analytic solution from SVD by simply retaining the top
r singular values (Eckart-Young-Mirsky theorem).
There are many applications based on low-rank matrix factorization because
in nature many things can be well-model as low-rank matrices.
If Louzhu you want to know more, you can learn something more on robust-PCA
which studies sparse+low-rank, which is more popular now I think. |
