Hello,
I am wondering why for the l1 constraint, can't we use the same formula as the one in the course but dividing by the l1-norm of the gradient? This would lead us to an offset delta satisfying the condition |delta|_1 <=1.
But doing the computations after, this effectively leads to a less small g(x+delta) than if we take (0,0,0,0,0,1). Why is it the case?
Q8/9 exam 2019
Hello,
I am wondering why for the l1 constraint, can't we use the same formula as the one in the course but dividing by the l1-norm of the gradient? This would lead us to an offset delta satisfying the condition |delta|_1 <=1.
But doing the computations after, this effectively leads to a less small g(x+delta) than if we take (0,0,0,0,0,1). Why is it the case?
Add comment