The Nesterov-Spokoiny Acceleration Achieves Strict $o(1/k^2)$ Convergence

Abstract: We study a variant of an accelerated gradient algorithm of Nesterov and Spokoiny. We call this algorithm the Nesterov--Spokoiny Acceleration (NSA). The NSA algorithm simultaneously satisfies the following property: For a smooth convex objective $f \in \mathscr{F}{L}^{\infty,1} (\mathbb{R}ⁿ⁾ $, the sequence ${ \mathbf{x}_k }{k \in \mathbb{N}}$ governed by NSA satisfies $ \limsup\limits_{k \to \infty } k² ( f (\mathbf{x}_k ) - f^* ) = 0 $, where $f^* > -\infty$ is the minimum of $f$.