Have you ever learned about Transfer Component Analysis(TCA) in transfer learning? It's key step is, $$\min\limits_W tr((W^TKHKW)W^T(I+\mu KLK)W)$$, which is equivalent to $$\max\limits_W tr((W^T(I+\mu KLK)W)^{-1}W^TKHKW)$$, and in the paper, it is said that similar to kernel fisher discriminant, the solution of $W$ is the eigenvectors corresponding to the $m$ leading eigenvalues of $(I+\mu KLK)^{-1}KHK$
So here are my questions, first is how the min becomes the max above?
The other is how to get the conclusion of the solution is the $m$ leading eigenvectors?
Either question, if you have an idea, please tell me.
The paper is https://www.cse.ust.hk/~qyang/Docs/2009/TCA.pdf