I want to find the minimum value of the objective function if we set K equal to the number of samples.

I know the objective function is $J=\Sigma_{n=1}^{N}\Sigma_{k=1}^{K}r_{nk}||x_n-\mu_k||^2$ And we take the derivative to get $\mu_k=\frac{\Sigma_{n=1}^{N}r_{nk}x_n}{\Sigma_{n=1}^{N}r_{nk}}$

However, when plugging that back in and set K=N, it seems unable to be simplified: $J=\Sigma_{n=1}^{N}\Sigma_{k=1}^{N}r_{nk}||x_n-\frac{\Sigma_{n=1}^{N}r_{nk}x_n}{\Sigma_{n=1}^{N}r_{nk}}||^2$

Any help?


Well, when k = n, the obvious global minimum occurs where $x_n = \mu_k$ for all cluster centers. This falls out from your derivate because $r_{nk} = 1$ iff $n = k$.

  • 1
    $\begingroup$ So would the minimum value then be 0? $\endgroup$
    – IrCa
    May 21 '19 at 22:38
  • 1
    $\begingroup$ Yes exactly. This is when every point is its own cluster center $\endgroup$
    – Sean Owen
    May 23 '19 at 2:26

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.