Problem understanding how a linear equation is simplified

30 Views Asked by At

Using this paper as a reference (Section IV.C, page 4318), We have the following objective function which we wish to minimize with respect to $D \in \mathbb R^{n \times K}$ ($X \in \mathbb R^{K \times N}$):

$$\min_{D} \{||Y-DX||_F^2\} \quad \text{subject to} \quad ||X||_0 \leq T_0$$

Which could be simplified as follow:

\begin{align} &\|Y-DX\|_F^2 \tag{1} \\ &\qquad = \left\|Y-\sum_{j=1}^K d_jx_T^j\right\|_F^2 \tag{2} \\ &\qquad =\left\|\big(Y-\sum_{j \neq k} d_jx_T^j\big)-d_kx_T^k\right\|_F^2 \tag{3} \\ &\qquad = \left\|E_k-d_kx_k^T\right\|_F^2 \tag{4} \end{align}

I'd like to make sure that I understand how we get $(4)$ from $(1)$.

Decomposing the multiplication $DX$ to the sum in $(2)$ is a matrix multiplication, right? The way I learned it, when you multiply two matrices, rows of the first matrix is multiplied by column of the second one. But what I see in the above formula is quite different. Then I guessed perhaps it gets the multiplication entries one by one, but how do we separate the kth column of $D$ multiplied by its corresponding row in $X$ ($d_kx_T^k$)? Please explain the simplification process in details because I need to understand it to be able to understand the whole topic of the above mentioned paper.

1

There are 1 best solutions below

4
On

Note that even though it looks confusing the sum is just the matrix multiplication: Vectors are multiplied with transposed vectors, resulting in a matrix. Those matrices are added. The result is the same as if you do the multiplication as usual. Example:

$$A = \Big(\matrix{a_{11} & a_{12} & a_{13} \\ a_{21} & a_{22} & a_{23} \\ a_{31} & a_{32} & a_{33}}\Big) $$ $$B = \Big(\matrix{b_{11} & b_{12} & b_{13} \\ b_{21} & b_{22} & b_{23} \\ b_{31} & b_{32} & b_{33}}\Big) $$

Then $a_1b_T^1 = \Big(\matrix{a_{11} \\ a_{21} \\ a_{31}}\Big) \Big(\matrix{b_{11} & b_{12} & b_{13}}\Big) = \Big(\matrix{a_{11}b_{11} & a_{11}b_{12} & a_{11}b_{13} \\ a_{21}b_{21} & a_{21}b_{22} & a_{21}b_{23} \\ a_{31}b_{31} & a_{31}b_{32} & a_{31}b_{33}}\Big)$.

The middle step should be clear, just one term in the sum is extracted.