3 problems

Let $F_m$ denote the objective of client $m$, let $F$ be the global objective, and let $H$, $B$, and $\sigma$ denote the smoothness, initial-distance, and noise parameters, respect…

Effectiveness of Local SGD. Local SGD can dominate mini-batch SGD over this problem class in the regime when $\tau, QB \mathbb{L} H$ and $\boldsymbol{\sigma},\zeta_\star$ are small…

The training loss is considered in the late phase of Local SGD, after it has decreased below the level where the reciprocal loss exhibits linear dependence on the total number of i…