In to solution to question 48 we seem to plug \( \gamma / \|w_*\|_2 \le M \) into the previous result \( t \le R^2 \|w_*\|_2^2 / \gamma^2 \) to get (
t \le R^2/M^2 ) . However in this last expression we are dividing \( R^2 \) by a larger value than before and therefore seem to be tightening the bound (constraining t to a smaller value than before). Why is it OK to do this?
Exam 2021 Question 48 Solution
In to solution to question 48 we seem to plug \( \gamma / \|w_*\|_2 \le M \) into the previous result \( t \le R^2 \|w_*\|_2^2 / \gamma^2 \) to get (
t \le R^2/M^2 ) . However in this last expression we are dividing \( R^2 \) by a larger value than before and therefore seem to be tightening the bound (constraining t to a smaller value than before). Why is it OK to do this?
1
same question. I spent much time to find a bound, i don't know why it is reasonable...
at the last part we have the constraint of \(||w||=1\) and \(yx^Tw \geq M \) which means that \(\gamma=M\)
1
Add comment