I think the correct 'tick marks' are missing accidentally everywhere. For most questions, this is fine, because you can see the 'comment' that was added. For the questions you mentioned, no comment was added. Would you like to discuss your answer to any of those questions?
I guess for problem 3, the lecture on least squares answers this question.
For problem 4, I was wondering if in the scenario where we want to minimize the l2 norm of each cluster center we pick as mu_k not the mean of the points in each cluster, but the minimum value of all x_n where z_nk = 1?
For problem 4, I wonder what an update rule should look like. In my opinion it should be the same as the whole algorithm of K-Means, except for when at the end of step 1 we minimize znk for the sum of Eclidean distance and l2 norm. Is that correct?
Exam 2016 solution incomplete?
There's no answer for question 3 and 4 for final exam 16. Is this intentional or is the solution missing by accident?
4
I think the correct 'tick marks' are missing accidentally everywhere. For most questions, this is fine, because you can see the 'comment' that was added. For the questions you mentioned, no comment was added. Would you like to discuss your answer to any of those questions?
I guess for problem 3, the lecture on least squares answers this question.
For problem 4, I was wondering if in the scenario where we want to minimize the l2 norm of each cluster center we pick as mu_k not the mean of the points in each cluster, but the minimum value of all x_n where z_nk = 1?
2
For problem 4, I wonder what an update rule should look like. In my opinion it should be the same as the whole algorithm of K-Means, except for when at the end of step 1 we minimize znk for the sum of Eclidean distance and l2 norm. Is that correct?
I sketched the solution this way but I'm not sure that I'm correct,
sorry for the bad quality of the picture (it's my old broken camera --')
Personally I have this:
Can a TA confirm it is correct?
8
Hi, why do you remove the L2 norm of (x_n - u_k) after computing the gradient ? Thanks for your help
Add comment