For the question 2 of the 2020 exam, it is said that the solution of SVM problems only depends on the support vectors. Just to be sure, we mean here essential support vector as well as bound support vectors? So if if change points that are within the margin, or just on the margin, it will change the SVM decision boundary?
For the question 8, I had the intuition that the uniform distribution was not belonging to the exponential family. But I don't understand the reason stated in the correction: "the supports depends on the parameters eta=(a,b)". Does that mean that if a function has an arbitrary support like [0,1] for which 0 and 1 are not parameters of eta, it can belong to an exponential family? Or exponential family functions always have infinite support?
Could you please explain more about the SVD part? For the essential support vector, if we set the subgredient of hinge loss h(x), with x=0, to be 0, why would SVD decision boundary be changed?
Question 2 and 8 exam 2020
Hello,
For the question 2 of the 2020 exam, it is said that the solution of SVM problems only depends on the support vectors. Just to be sure, we mean here essential support vector as well as bound support vectors? So if if change points that are within the margin, or just on the margin, it will change the SVM decision boundary?
For the question 8, I had the intuition that the uniform distribution was not belonging to the exponential family. But I don't understand the reason stated in the correction: "the supports depends on the parameters eta=(a,b)". Does that mean that if a function has an arbitrary support like [0,1] for which 0 and 1 are not parameters of eta, it can belong to an exponential family? Or exponential family functions always have infinite support?
Hi,
About SVMs: yes, this sounds correct.
On exponential families: the support of an exponential family must remain the same across all parameter settings in the family.
There are multiple correct arguments why this example is not part of the exponential family, this is just one.
1
Thank you so much!
Could you please explain more about the SVD part? For the essential support vector, if we set the subgredient of hinge loss h(x), with x=0, to be 0, why would SVD decision boundary be changed?
1
Add comment