Is it False because we don't need "a bounded and convex constraint set"? In the lecture, properties of the set are indeed not mentioned (only \(X \subseteq \mathbb{R}^n\)) but why we work in a so wide setting where X can be the whole space and then Frank-Wolfe is unable to do even one step? Should we have at least some restrictions on the set?

## Exam 2020 Q18

Hi,

Is it False because we don't need "a bounded and convex constraint set"? In the lecture, properties of the set are indeed not mentioned (only \(X \subseteq \mathbb{R}^n\)) but why we work in a so wide setting where X can be the whole space and then Frank-Wolfe is unable to do even one step? Should we have at least some restrictions on the set?

The problem miss the statement that s=LMO(gradient f).

Isn't it because it should be O(1/T) and not O(1/t)?

## 2

## Add comment