0
$\begingroup$

I have the following optimization problem(objective function is non-convex. B and c are unknown variables) $$minimize_{B,C}\sum_{n=1}^N{\Vert\sum_n -\mathbf{B} diag(\mathbf{c}_{n})B^T \Vert_F^2} $$ subject to $$\Vert \mathbf{b}_k \Vert_1 \le \lambda , k=1,...,K $$ $$-1\le\mathbf{b}_k(i)\le1,$$ $$max\vert \mathbf{b}_k(i)\vert =1, i=1,...,P$$ $$\mathbf{c}_n\ge0 , n=1,...,N$$ I want solve it with projected gradient descent and implement with matlab , but it has multiple constraints and I don't know how solve it!!! please help me...

$\endgroup$
0
$\begingroup$

If both $ B $ and $ {c}_{n} $ are to be optimized a simple solution would be to chose $ {c}_{n} = 0 $. Then just chose one element of $ B $ to be 1 or -1.

This will yield cost value of $ 0 $ which for this problem is a lower boundary.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.