2
$\begingroup$

Focusing on following least squares problem: $$\min\limits_{V} \lVert Z - WV \rVert _{_F}^2$$

$$Z∈{R}^{m*n},\quad W∈{R}^{m*k},\quad V∈{R}^{k*n},\quad k\lt m\lt n $$ This problem can be easily solved given $Z$ and $W$. I assume the solution of this problem as $\overline{V}$.

My question is that, how can we recover $W$ by $\overline{V}$ and $Z$?

I found something called Inverse Optimization, but cannot figure out the relationship between them.

EDIT: It is acceptable to have a set of $W$ or one feasible $W$ if $W$ is not unique.

($\overline{V}$ and $Z$ are both nonzero matrices, and $Z$ is not promised to be full rank).

$\endgroup$
6
  • $\begingroup$ Whatever your W was fixed at in the least squares problem in which you solved for V. Of course, this combination of V and W won't even be a local minimum, let alone a global minimum, if convergence of the overall algorithm has not yet occurred. scicomp.stackexchange.com is probably a better location for this question. $\endgroup$ Commented Jun 5, 2018 at 1:52
  • $\begingroup$ @MarkL.Stone Thank you. You are right. What I concern is an intermediate result before convergence. $\endgroup$
    – lee
    Commented Jun 5, 2018 at 2:05
  • $\begingroup$ Well, intermediate results may be lousy. If it hasn't converged to a global minimum,. even the converged result could be lousy. $\endgroup$ Commented Jun 5, 2018 at 2:09
  • 1
    $\begingroup$ cross-posted $\endgroup$ Commented Jun 5, 2018 at 13:12
  • 1
    $\begingroup$ In general the answer is not unique. Let's say $Z=\bar V=0$. Then $W$ can be absolutely anything. You need to assume some non-degeneracy to make it meaningful. So, what are the assumptions you are operating under? $\endgroup$
    – fedja
    Commented Jun 5, 2018 at 15:37

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.