Consider a convex function f(𝐱)f(\mathbf{x}) bounded between [B,B][-B,B] on a constraint set 𝒮\mathcal{S}.

The center-of-gravity method finds 𝐱ˆ\hat{\mathbf{x}} satisfying f(𝐱ˆ)min𝐱𝒮f(𝐱)+ϵf(\hat{\mathbf{x}}) ≤ \min_{\mathbf{x} \in \mathcal{S}} f(\mathbf{x}) + \epsilon using O(dlog(B/ϵ))O(d \log(B/\epsilon)) calls to a function and gradient oracle for convex ff.

#incomplete


see: convex optimization


References:

  1. https://cims.nyu.edu/~cfgranda/pages/OBDA_fall17/notes/convex_optimization.pdf